Contents
Feature Special
Guest editorial, Jan A. Audestad ...... 1 Voice quality under ATM cell loss, Jens Berger and Mechthild Stoer ...... 85 Telecommunications and Complexity, Jan A. Audestad ...... 2 Traffic from Mobility in Mobile Broadband Systems, Fernando J. Velez and Luis M. Correia ...... 95 The role of uncertainty in strategic planning, Stein W. Wallace ...... 21 The haves and have-nots – The distribution of ICTs in Norwegian households, Rich Ling ...... 102 Strategy for Enterprise Specifications, Jan A. Audestad ...... 24
Agent Nets: a methodology for evaluation of enterprise strategies in information economy, Alexei A. Gaivoronski ...... 33
Technological Evolution: Some Strategic Considerations, Jan A. Audestad ...... 49 Kaleidoscope
An Object Oriented Model of the NORDTEL – 80 years of Nordic Telecommunications Business, Jan A Audestad, co-operation in the telecommunications area, Arvid Kjæreng and Laurent Mahieu ...... 54 Einar Utvik and Villy Toft ...... 109
Real Options and Managerial Flexibility, Stein-Erik Fleten, Trond Jørgensen and Stein W. Wallace ...... 62
New Aspects of Service Provision and Technology Strategies in Telecommunications, Shane Dye, Asgeir Tomasgard and Stein W. Wallace ...... 67
How to Survive in the Future, Jan A. Audestad ...... 74
Editorial office: Telektronikk Telektronikk Telenor AS, Telenor Research & Development P.O. Box 83 N-2007 Kjeller, Norway Volume 94 No. 3/4 – 1998 Tel: (+ 47) 63 84 84 00 ISSN 0085-7130 Fax: (+ 47) 63 81 00 76 e-mail: [email protected] Editor: Ola Espvik Editorial board: Tel: (+ 47) 63 84 88 83 Ole P Håkonsen, Senior Executive Vice President Oddvar Hesjedal, Vice President, Research & Development Status section editor: Bjørn Løken, Director Per Hjalmar Lehne Graphic design: Tel: (+ 47) 63 84 88 26 Design Consult AS Editorial assistant: Layout and illustrations: Gunhild Luke Gunhild Luke, Britt Kjus, Åse Aardal Tel: (+ 47) 63 84 86 52 Telenor Research & Development Editorial
JAN A AUDESTAD
Strategy has to do with the way In Telenor (or rather Televerket) organisations or firms are man- operation research began during aged. Strategy as a process is not the 1970s. One key problem then the management of the organisa- was to use linear programming tion itself but rather the way in methods to find locations for which management obtains infor- exchanges and other equipment in mation in order to make decisions. order to make the network as Strategy and tactics are terms refer- cheap as possible. These are ring to the same thing and consist problems which are still with us. of methods supporting a common goal, namely the ability of a com- Since the early 1990s the com- pany or organisation to achieve its plexity of telecommunications has goals: in warfare to win the battle; increased enormously. This has to in industry to seize the market. do with several independent forms Tactics has to do with how strate- of evolution: evolution of new ser- gic knowledge is put into opera- vices in the basic portfolio of tion. In this meaning, strategy is telecommunications, application concerned with the long term plan- of telecommunications to new ning and manoeuvrings of the com- areas, evolution of the technology pany. toward processing making tech- nologies cheaper and more versa- Encyclopaedia Britannica defines tile, liberalisation of the market the difference between strategy and opening for competition and tactics as follows (in the Macro- new organisation of the business, paedia under Industrial Engineer- and regulation of the competition. ing and Production Management): (1) the longer the effect of a deci- As a result of the technological sion and the less reversible it is, the development society has become more strategic it is; (2) the larger the proportion of a system that is more dependent on telecommunications. The evolution is thus affected by a decision, the more strategic it is; and (3) the more something which concerns not only the telecommunications concerned a decision is with the selection of goals and objectives, industry but also society as a whole. The study of the impact of as well as the means by which they are to be obtained, the more this evolution on society has not really begun but is likely to strategic it is. In this definition it is tacitly understood that the ref- become a major subject taught at universities in the near future. erence is a continuous scale ranging from tactical to strategic deci- This can also be assessed from the popularity gained recently by sions. The scale is not only dependent on time – where tactical is the conferences of the International Telecommunication Society. short term and strategy is long term – but also on the quality of the These conferences are not concerned with technology but with decision: degree of irreversibility, part of the system affected and policies and interactions with society. Inspired by this way of the number of goals concerned. looking at the telecommunications industry, we have started this collection of papers with a discussion of the complexity of Below we are not concerned with the strategies themselves but telecommunications, where we try to evaluate several of these with the methods that can be applied to identify and elucidate complexities and describe how they interact. The intention is to strategic problems. These methods are referred to as operation show that these problems cannot be understood unless proper research. Operation research is mainly concerned with methods methods are used to analyse them. supporting the management of industrial organisations, govern- ments and defence systems, and is as such a multidisciplinary sci- The remaining papers are concerned with different aspects of the ence drawing on mathematical and statistical methods as well as strategic problems of the industry. These problems can only be new areas such as cybernetics, behavioural sciences, organisation understood in terms of the methods of operation research. The theory, decision theory, game theory and nonequilibrium econ- papers do not provide a complete picture of this vast subject mat- omy. ter but hopefully they will encourage further research of the strate- gic problems of telecommunications. Such studies have been done Operation research is a rather young science. Though operation with success for the oil industry and the energy industry. The research can be regarded as the science of applying scientific much more complex telecommunications industry has not yet methods to management and as such has been with us for a long been subject to this type of analysis in depth. This is an interna- time, it is usually agreed that its modern evolution as a separate tional phenomenon, and the education in business management of study began in Britain in 1937 when scientists were led to teach our own managers to be is based on studies of other and compara- military leaders how to use radar in anti-aircraft warfare. The tively much simpler business systems. methods – or rather the selection of methods – were developed during World War II, and several operation research groups were The present papers constitute one of the first collections of established analysing the impact of different ways of conducting operation research studies for the telecommunications industry mine warfare, air raids and ground battles. The first scientific ever published! groups dealing with this new subject were formed around 1950, much because of the success of the method during the war. How- ever, the methods were not applied at any significant scale for industrial management until the late 1950s.
Telektronikk 3/4.1998 1 Telecommunications and Complexity
JAN A. AUDESTAD
1 Problems the effect of the change may come two-fold. First, quality of service is bar- rather slowly. One example is again gained against low price as for the Inter- The main problems we are facing in the liberalisation of the market, which net, and the customers – residential or telecommunication are all related to com- represents a major change in one of the business – are sensitive to price but have plexity: variables, where it may take several little awareness concerning reliability. years before the impact is really seen Second, the way in which administrative • Complexity of technology; on the incumbent (Sweden and the data systems are specified and imple- • Complexity of products and markets; United Kingdom are examples of this). mented is different from that of telecom- Having an expectation that damping munications equipment: in the latter, • Complexity of organisation; does not exist or is insignificant, may most of the effort is put in handling of • Complexity of co-operation or com- lead to wrong regulations of the mar- exception conditions, autonomy, auto- petition; ket; matic fault recovery and fault isolation. Telecommunications systems have also • Complexity of value creation. • They may be described in terms of a strong real-time and concurrency require- large number of variables; that is, the ments. The real-time requirement implies This complexity is new. It did not exist spatiotemporal problem has too many that all processing must be done within a decade ago. Then the telecommunica- dimensions for simple visualisation of strict time limits. These considerations tions industry consisted mainly of it. Chapter 8, where the variables are almost absent in administrative sys- monopolies or cartels, the services were required for describing the business of tems. Solution of real-time problems few and simple – and the Internet did not providing access, is an example of this; require use of dedicated modelling exist outside the academia and the re- • They may, from time to time, show methods for telecommunications soft- search institutes. The companies offered chaotic behaviour so that the future ware reflecting the dynamic behaviour everything to everyone within their geo- development may be impossible to of the systems. Concurrency means that graphic domain. When competition predict; thousands of simultaneous usages of the came, it came at a time of transition. This telecommunications software may exist was a transition: from person-to-person • They are not in general in an equilib- in the same exchange. Because of the communication to distribution of content, rium state. real-time requirements this implies that from vertical and tight organisation of each usage requires a separate copy of the telecommunications companies to Simple economic analysis cannot be used each software module and that the execu- looser organisation and subdivision and in order to describe such systems. On the tion of these copies are not synchronised. internal competition, and from national other hand, advanced analytical methods These considerations make the design of and localised industries to multinational such as game theory, dynamic control telecommunication software a separate – conglomerates forcing the telecommuni- theory and optimisation techniques do and complex – discipline of computer cations industry to form international not contain all the tools required for science. alliances. All this has led to new and analysing such complex situations. This poorly understood strategic complexities. even applies to models based on com- This may have an impact on the conver- puter simulation. Therefore, the problem gence of telecommunications technolo- The complexity depends on many vari- has to be approached from different gies and information technologies making ables and varies over time, or has spa- angles and at different levels of granu- this convergence not so trivial. Moreover, tiotemporal behaviour as it is defined in larity, where the first objective should be the data industry is entering the arena of the mathematical literature [1]. Such to understand the underlying complexity the telecommunications industry produc- dynamic systems may have several and its dynamics. This is the main goal of ing equipment for switching of traffic. characteristics which make them difficult this essay. Since these two industries have different to analyse: production philosophies, this may change The dependability of society on telecom- • They are generally non-linear; fundamental requirements of telecommu- munications is becoming more critical. nications on reliability making the net- • They will contain both fast and slow Telecommunications installations may work cheaper but less dependable. This fluctuations at the same time, where become the target for terrorism, not only aspect is probably one of the most impor- some fluctuations may even be faster by physical attack but also using logical tant elements of a technological strategy than the relaxation time1 of the sys- bombs and other ‘soft’ means. This is an for telecommunications. tem; aspect which has not been fully under- stood by politicians and regulators. Com- • They may be subject to large external The markets are also changing. Most of petition in a market with shrinking mar- disturbances. Liberalisation of the the services enhancing the basic tele- gins may cause reliability to be bargained telecommunication market is such a phone service are new. They became uni- against lower prices. This is one lesson disturbance changing the basic model versally available less than ten years ago. learned from the Internet: much of the of the system; The age of services requiring the intelli- success is related to low prices and high gent network technology is less than five • They may also be resistant to sudden functionality. However, reliability and years. These developments are the basis changes – or be overdamped, that is, quality of service is low. for the distribution of content rich ser- 1 The relaxation time is a measure for vices. Since we have had these services Another experience is that the quality of the time required for a system being for such a short time, it is very difficult service in terms of reliability, response brought out of equilibrium to again to estimate how they will evolve in the time and stability is poor also for local settle in the same (or a new) equilib- future: they have no history to base fore- data networks. The reasons for this are rium state if it is left alone. casts on, the technology on which they
2 Telektronikk 3/4.1998 are built are evolving and changing time been reduced significantly during offering protection against fraud, simple rapidly, and telecommunications has the last few years? The surprising answer handling of service menus, and memory become a competitor to other distribution is that this is not so in general, though of long dialling sequences, the service is channels of many products (newspaper, there are systems and methods which do difficult to use. What is required is not a trade and so on) and it is difficult to just that in particular circumstances. This better IN platform but a new platform assess the persistence of traditional chan- is the core of the problem we are facing replacing the telephone apparatus at nels: a change will come for certain, but today: there are two paths of develop- every home and industry by new equip- the problem is when. ment of telecommunications where one is ment offering simple interfaces based on slow and the other is fast. The fast de- smart cards, displays and automatic han- The liberalisation of the telecommunica- velopments are associated with new dling of complex menus. On the other tions market is changing the way in products on the World Wide Web and hand, GSM offers just this type of inter- which the business is organised and con- new services on intelligent network face, and because of its widespread use it ducted. The monopolies were vertically nodes; that is, development of products is replacing the need for UPT. organised companies offering all aspects which can be made from modules and of telecommunications to everyone. tools existing on the platform on which it Let us now review some of the major Competition is now met in small niches is implemented, such as html on the events leading up to telecommunications of the business as well as from vertically Internet and service independent building as we perceive it today. integrated companies mimicking the blocks (SIBs) on the IN. The develop- incumbent. This is making the industry ment of the platform itself takes time, The first major event in telecommunica- complex and requires a new look at how and is an example of the slow path of tions was that Strowger invented auto- value is really created within the indu- development. matic switching in 1889, leading to the stry. In the core business the markets first commercial electro-mechanical have been big with good margins. This is As an example, it took more than ten switch in 1898. Although the technology changing since competition has forced years to develop the IN technology; in offers cost reduction for the operators, it price reductions on core products. Many contrast, it takes only days to develop a took 75 years to automate Norway com- of the new markets have not materialised service like freephone on this platform. pletely. And Norway is no exception in yet. This leaves the industry facing a This duality is becoming a strategic this evolution. shrinking primary market and a new problem in telecommunications: it has market with much potential but high changed our expectation concerning Almost forty years after automation of investment risks. the time it takes to bring a new product telephone switching the first automatic to the market, and it causes us to under- systems for transfer of text were in- The new situation requires that the estimate the cost of radically new prod- vented. In 1933 we got the first telex net- telecommunications operators are recon- ucts as compared to products which work in Germany, though this service did sidering how value is created in the can be designed on ‘smart’ platforms. not become generally available until after organisation and when it is advantageous We also often overlook the fact that there World War II. During the 1960s the first to co-operate or compete with newcom- is a continuous evolution of these ‘smart’ communication satellites were put into ers in the marketplace. These are difficult platforms, which means that the two orbit, experiments with data transmission problems and may require development development paths often coexist. This began, and the first data controlled of new strategic methods. means that we also often have the choice exchanges were implemented. The back- of either implementing a service or appli- ground invention enabling this radical These problems are considered below, cation on the existing platform, or wait evolution was the transistor. These sys- where the main concern will be to assess until a new version of it is in place. In tems required high reliability and small the complexity all these items will the first case we may develop a service size which could not be achieved with impose on the industry. Access operation which is not optimal for the market; it the vacuum tube. will be used as an example for illustrat- may even be detrimental to it. In the sec- ing how big and complex a strategic ond case the product may come too late About 1980 we got the first automatic analysis is, even when it is confined only to the market because other developers mobile systems. These included NMT, to identifying the parameters of one par- have come up with replacement products, AMPS and TACS2 for land mobile appli- ticular part of the industry. or the market has changed the focus cations, and INMARSAT3 for communi- towards other products. 2 NMT (Nordic Mobile Telephone) is the 2 Complexity and Take the universal personal telecommu- Nordic automatic land mobile system Technology: Does new nications (UPT) service as an example of operational from 1980. AMPS (Auto- this problem of duality. The service is matic Mobile Phone System) is an Technology Reduce marketed in Norway as ‘Alfanummer’. American land mobile system put into the Time to Market? The idea behind the service is indeed a operation during the first half of the smart one: the service offers mobility 1980s. TACS (Total Access Control The evolution of telecommunications has between any termination – fixed or System) is a British adaptation of been rather slow but is accelerating. By mobile – in the network on one number AMPS put into operation in 1983. slow we mean that the time between and one service profile of the user. The major events and the time to invent new service is easily – and cheaply – imple- 3 International Maritime Satellite Orga- technologies have been long, in general mented on the IN platform. However, nization inherited the MARISAT sys- more than ten years. One important issue since this platform does not offer a man- tem from Comsat General and com- is then to answer the question: has this machine interface to the user capable of menced operation in 1980.
Telektronikk 3/4.1998 3 NC 5 % transmission Java Complexity 95 % processing Agents 95 % transmission Web 5 % processing IN GSMT Land mobile INMARSAT Computer controlled switching Automatic Data transmission switching Telex Satellites
1898 1930 1960 1980 1990 1995
Transmission Processing
Applications Figure 1 Evolution of telecommunications
cations with ships. None of these tech- In Figure 1 the change of focus is illus- processor. The technology of echo can- nologies would have been possible with- trated by how much of the specifications cellation has during these twenty years out the microprocessor. After 1990 we of the INMARSAT system and of the developed from pure hardware construc- have seen an explosion of new technolo- GSM system were concerned with trans- tion to pure software construction. This gies: intelligent networks, GSM, World mission and processing. During the example is typical for much of the evolu- Wide Web, network computer, Java, decade the focus changed completely. tion which has taken place during the last intelligent agents, value added services, The change of focus has also resulted in twenty years. In addition, the use of soft- and value added networks. It may look as a tremendous increase in complexity dur- ware technology has increased the num- if the complexity of new systems ing this short period of time. The ber of applications which can be con- invented at any time is proportional to increase in complexity of the systems is structed. the complexity of the previous systems owing to the increased capabilities of the taken together, giving rise to an evolution microprocessor: procedures and algo- The focus has again shifted during the which is exponential – often referred to rithms that had to be implemented in 1990s. Because of the web, the designer as natural evolution. This evolution is hardware are now implemented in soft- may now turn the main focus towards the illustrated in Figure 1. ware, and procedures that could not be applications and less towards the technol- implemented at all or were highly im- ogy enabling them. Until about 1980 the main focus in practical are now possible as computer telecommunications was on transmission. solutions because of the low cost of However, it should be noted that de- After 1980 the focus has been on pro- microchip computers, advanced develop- velopments that involve new software cessing. By processing we mean acti- ment tools for software production, and methods or hardware still take time – vities related to implementation of both more processing and storing capacity per from five to ten years or more – so that services, applications, protocols and plat- unit of area. One example is the echo the development time of such solutions forms, and complex procedures like canceller used to suppress echoes re- have not been reduced significantly. This cryptography, channel coding, voice cod- turned from the terminations of the four- applies, for example, to the development ing and error protection coding. In wire telephone network. In 1970 these of nodes offering intelligent network (IN) advanced products like GSM, demodula- devices had to be about one cubic metre capabilities and enhanced user capabili- tion of radio signals, voice coding and big in order to accommodate the lengths ties (CTI4), software platforms like the channel coding are specified and imple- of coaxial cables and the number of mag- World Wide Web and CORBA5, and mented as processing activities. In 1986 netic switches required for constructing new protocols such as IP66. However, the first simulations of channel coding the transversal filters of the canceller. applications built on platforms already and demodulation of GSM were per- Ten years later this bulk was reduced to a implemented such as intelligent network formed on a Cray computer. This was small printed circuit card containing a nodes, the World Wide Web and the only machine at that time fast enough few microcircuits and a few other com- CORBA can often be developed in a to do this using software only. Five years ponents. Ten years later still the whole matter of days or, at most, during a few later the processes were easily imple- device was contained on one microchip weeks. This has created an imbalance in mented on microchips. not distinguishable from any other micro- the development time of new products:
4 Telektronikk 3/4.1998 Platform products Development time Technology is thus an example of an evolutionary variable in the dynamics of Intelligent network (IN) node 10 years telecommunications containing two (or GSM 10 years more) types of temporal cycles: short World Wide Web 5 years cycles and long cycles. This may also CORBA 10 years give rise to different behaviour of com- ISDN 15 years peting companies depending on where Computer telephone integration (CTI) 10 years they are in these cycles. One example is that companies focusing mainly on the Derived products Internet may be in a period with short Freephone 1 month cycles concerning the development time Universal personal telecommunications (UPT) 6 months for new products. Other companies may Premium rate service 1 month be in a period of predominantly long Web page 1 week cycles building up the technology which Electronic newspaper 2 months may make them change to short cycles in Transaction services 2 months order to compete with the Internet. This Products with enhanced technology is illustrated in Figure 3. The profits may as a result also be cyclic and may corre- Centralised queue 1 - 2 years spond to a special kind of leapfrogging Intelligent building 1 - 2 years [2] which may be common in telecom- Wearables: experiments months munications in the future. The compe- Werables: product adapted to market several years tition between such companies is not equal – and cannot be made so by any Figure 2 Development time of different technologies regulatory interference.
If we look at the technological evolution we may divide it into the following main phases [3], [4]. Before 1980 the focus those which can be developed on such The price tag will be high for such solu- was chiefly on transmission. This was the platforms and those which cannot. The tions, while it takes time to develop the great era of multiplexing technologies first type of product will have a short cheap technology. and source coding techniques for both development time and, hence, a short speech and picture. The 1980s were the time to market, while the second type of These elements are summarised in Figure years of processing where the micropro- product will have a much longer time to 2. cessor led to the PC revolution and en- market. In many cases this leads to un- abled the design of equipment for mobile realistic expectations concerning pro- duction time and resources required to implement a given solution. This applies in particular to products where most of the product can be made on such a plat- form but some part of it needs to be de- veloped. One example is the implementa- Company one tion of the centralised queue service which can be implemented on an IN node in a very short time but requires new development involving remote procedure calls between the PABXs and the IN Profit node. For the intelligent building the situ- ation is similar: the product can be imple- mented quickly on existing platforms.
4 CTI is equipment combining telecom- Company two munications and computer applica- Developmen tions such as deriving customer infor- cycles mation from the called number. It is an abbreviation for Computer Telephone Integration. 5 CORBA is an abbreviation for Com- mon Object Request Broker Architec- ture, and is a new industrial standard for platforms for distributed processing. 6 Internet Protocol version 6. Figure 3 Product cycles
Telektronikk 3/4.1998 5 communications and control devices in now 180 main switching nodes in the A measure for dependability is suggested automobiles. The PC did not lead to telephone/ISDN network. One of the in [6]: more traffic in the telecommunications GSM networks in Norway has about D = α log(NT), networks. The PCs wre isolated ma- 40 % as many users as in the fixed net- chines in the homes or connected to work. These are, however, served by where the measure for dependability is closed networks in companies. The only eight exchanges. What has D, α is some constant of proportionality, 1990s have been the era of access to happened is that each node has become N is the number of customer circuits information manifested by the World larger, and a failure of any such node will affected and T is the total downtime in Wide Web and the search engines. The affect larger parts of society and require hours. If we put α = 1, this measure is enormous increase in traffic caused by an increase in reliability of each node of in many ways equivalent to the Richter the Web came about because the PCs the order of one or two in magnitude in scale for earthquakes (however, do not were there as a result of the previous order to keep the risk level at a constant bring the analogy too far): development. According to [3] the next value. In addition comes the general • If the value is less than 6.0 the impact decade will be the era of interaction increase in capacity of optical fibre sys- may be severe but not critical. This where micro-miniaturised sensors and tems (order of terabits/s) creating a net- will happen if one million circuits are remotely controlled devices will be the work with overwhelming overcapacity affected for one hour, or 100,000 cir- basic components. One example is the where there is little concern in bringing cuits are affected for ten hours; video camera now reduced to a calls over long distances and thus wast- microchip with the micro-miniature lens ing capacity. This means that we can • If the value is above 7.0, the impact glued to or integrated within the chip. reduce the number of switching nodes may, just as for earthquakes, be dis- The price is about $ 10, and will be by connecting users over long distances astrous. [6] estimates that the outages reduced significantly during the next few to the few nodes which are left. in the USA in the summer of 1991 years. Similarly this development may were of this order. enable micro-miniaturised radars, chro- In this way the network may be made matographs, pumps, valves, motors and smaller with regard to the amount of Using a value of α = 1 may apply for so on, revolutionising many current tech- equipment required. However, this may very large networks such as the whole nologies such as medicine, environmen- not result in a proportional reduction in international network. A larger value of tal control, process management and sur- cost since the reliability must be in- α may be used to estimate the impact on veillance [5]. This technology will creased in order to keep the effects on smaller networks. However, we should indeed require telecommunications! society small. One good example is the not put too much significance into this But telecommunications will most likely cost of satellites. They are a comparably formula: it is merely included here to change in order to meet new require- simple equipment, but the cost is tremen- visualise the general problem of depend- ments with regard to bandwidth, ubiq- dous because of the reliability require- ability. uity, pricing, and so on. ments which must take into account the harsh environment and their inaccessi- Values above 7.0 are rare but their proba- Of all these phases, the telecommuni- bility to repair. bility may be increasing for several rea- cations industry has only been in the sons: driver’s seat for one of them: the de- The sensor technology of the next decade • The increased complexity of telecom- velopment of transmission systems. may require new usage of telecommuni- munications networks; The other areas have come as a result cations in other areas of society increas- of developments in the data industry, ing the dependability between telecom- • The tendency to centralise nodes as the information industry, the microelec- munications and vital functions of indu- explained above; tronics industry and the academia. This stry, governments and individuals. • The increased need for frequent up- is also what makes the telecommunica- grades of software and hardware of tion industry so vulnerable: its field may The dependability of the society on nodes in order to meet market de- easily be invaded by companies skilled telecommunications is increasing and mands for new services and reduce in these new areas. Still we are struggling may already be critical in certain areas, operation costs; with the convergence of telecommunica- amongst others the impact on our eco- tions and information technology while nomic systems. One unanswered ques- • The reduced capital available for the new development may require other tion is how long the New York Stock investments and maintenance because types of convergence completely un- Exchange and the Tokyo Stock Exchange of the liberalisation of the market known to us. can be without telecommunications resulting in competition between oper- before the world enters into a severe eco- ators bargaining reliability for price. 3 Complexity and nomic crisis. Another question is how quickly the air transport systems of the Another problem is related to the number Dependability: The world will break down as a result of fail- of error reports created by the network. Impact on Society ure of a few critical nodes. Will telecom- In a network with n nodes where all the munications become the new targets for nodes are connected with one another the terrorists in order to hurt the society? number of error events generated if one The number of main switching nodes in Will they use physical attacks or timed node fails is also n. With a mean time the telecommunications network has logical bombs? Will such bombs be between failures of a node of t in years been reduced considerably during the last implanted at the time of construction or the mean time between failure of the sys- ten years, and there is still much potential inserted later via insecure accesses to the tem will be t / n, and the number of error for further reduction. In Norway there is network’s operation systems? reports created per year will be n2 / t.
6 Telektronikk 3/4.1998 With a mean time between failures of 1 Ternary relationships: year, the number of reports created in a • vendor network consisting of 1,000 nodes will • external policy Ternary roles: be one million per year, or about 2,700 • regulation • product enablers per day. In a large network consisting of • binding 100,000 nodes, the number of reports per day will be about 27 million. Even if the Secondary roles: connectivity of the network is only 1 %, • product components the number of reports per day will be Secondary relationships: 270,000 in such a big network. The latter • production Enterprise number should indeed be taken seriously • regulation • primary product when considering the reliability of the • commercial Internet which already consists of a large • cooperation number (several millions) of nodes • policies (servers) with rather low reliability. Figure 4 Product hierarchy Hence, we see that the reliability of the network may become a cost driver since every error occurring in the network requires maintenance activities. In order a new product depends not on whether it telephony, delivery of electronic news- to reduce the number of points in the net- is done on a telecommunications plat- papers and advertisements is another work requiring maintenance, the number form or on an Internet platform. The dif- example. Electricity, cable TV and tele- of nodes should be made smaller, that ference lies in whether all components r- phony may also be combined as a single is, increasing the size of each node. In equired for the product are available at offer and on one bill. Services may be accordance with the above consideration the platform or need to be developed delivered by one company or be com- this may increase the dependability of first. In the latter case, the development posed of services delivered by several society on the network so that the two time and production cost are more or less specialised companies. Alliances be- strategies may be in conflict with each the same for the two types of platform. tween different actors may be built on other. There may, however, be a qualitative dif- single service offers or on portfolios of ference between the products. We will services and other products, resulting in come back to this in Chapter 7. interesting and difficult games of import- 4 Complexity and ance, dominance and buy-outs. This type Services: New Ways Services are just as much bundling and of complexity is treated in [7]. combination of offers as single products. of Making Products One example is bundling of TV channels, Telecommunications services for the local advertisements and subtitles over a business market may be used by the com- Until the beginning of the 1990s, services satellite network involving several in- panies purchasing them as ‘raw’ material comprised mainly simple two-party calls. dependent business systems. Electronic or as components for building their own The main focus was the residential mar- trade combined with payment services, products. In this way telecommunications ket. Since then we have got supplement- security and anonymisation services, may become a secondary or even ternary ary services such as call forwarding, call physical delivery of goods, delivery of product in this production. Electronic barring and call waiting and virtual pri- vate networks, and, most important, the Web on the Internet. The Web provides us with a completely new concept of what services are. The network com- puter, the intelligent agent and the Java language are derivatives of this technol- Ternary products: ogy. Note that the development of Ternary relationships: • telecommunications • remote interactions advanced telecommunications services • regulation of competition • market regulation • banking and the derivative technologies of the • money transfer Internet took place at the same time. This has created two strategic problems in Service provider products: telecommunications. First, it seems as if • order system developing solutions using Internet tech- • transactions • delivery of goods nologies is much faster than using Secondary relationships: • storage • enabling services telecommunication platforms, and sec- • management of logistics ond, that the products derived from the Enterprise for Internet platform are cheaper and better electronic commerce: than the equivalent products derived • branding • logistics management from telecommunications technologies. • alliances This may be true in special cases but is not true in general. As was shown above, the difference in the time to manufacture Figure 5 Product hierarchy: electronic commerce
Telektronikk 3/4.1998 7 banking, electronic newspaper and elec- Service tronic commerce are examples where providers Distributing contents telecommunications services are just ena- Internet blers for the primary service. The tele- Premium communications product may enter the value producing system of the company Telecommunications ContXt as an element of or as a support activity operator Xpert in a value chain, as a layer in a value net- U&I work or as a skill in a value shop [8]. The Connecting Market+ product is different in each of these cases subscribers because it will meet different interfaces. The impact of this is poorly understood by the telecommunication operators and is probably critical to them. In this mar- Figure 6 Interaction with service providers ket it is difficult to analyse what the product really is and how it is viewed by the buyer, and how it will evolve.
Figure 4 is an enterprise model where it is shown how a three stage hierarchy of systems). In this example telecommuni- operator. The traditional business of ‘influences’ may be incorporated in the cations is a ternary product. If the goods telecommunications is to interconnect production chain of the enterprise under were delivered electronically (music, subscribers, and the main product has consideration (the inner oval in the fig- video, pictures), telecommunications been to offer two-party calls between ure). The principles are equivalent to may have been upgraded to a secondary people and between machines. In the new those defined for distributed systems [9] product in the delivery chain. In this and market the business idea is to use but slightly modified in order to be appli- similar examples it will never be up- telecommunications for distribution of cable to systems which are more general graded to become a primary product. content. The telecommunications prod- than distributed computer systems (see ucts may then enter the production chain also [10]). of the customer as secondary or ternary products as explained above. The The enterprise produces a primary equip- 5 Complexity and telecommunications operator will see the ment for the market. This product may Market: New Ways content as emanating from the service require resources and elements obtained provider (in the figure indicated by ficti- from secondary sources. The product of Doing Business tious names Internet Premium, Xpert and may also be directly influenced by regu- Another way of looking at telecommuni- so on). For the service provider telecom- lation and other influences external to the cations products is shown in Figure 6. munications is just one way of distribut- enterprise. The ternary relationships There are two markets: one market where ing the content in order to reduce cost operate on the secondary elements of the telecommunications is the primary prod- and reach new customers, and it is only product and may not be visible to the uct, and one market in which the secondary to the main product which is enterprise. Still they may influence the telecommunications operator distributes the content. The main value the telecom- product. contents from service providers to people munications operator can offer in this and machines connected to the network configuration is the number of sub- Figure 5 illustrates these principles for operated by the telecommunications scribers connected to the network who the electronic commerce. The product components provided by the primary enterprise may be branding, logistics and building of alliances. Service providers deliver secondary products such as order systems, transaction support and delivery Service of goods. The business of the service providers providers may then be influenced by the Internet Levi's Premium regulation of the competition, make use of the banking system, and use telecom- ContXt Coca Cola munications as an enabler both for the Xpert Times remote ordering system and the transac- U&I MIT tion support system. For enterprises liv- Market+ Naxos ing from the electronic commerce it is Content not important how this is done. For the providers enterprise it is more important that pay- ment transactions are managed. It is not Telecommunications important to know how and by whom operator this is done. This is exactly the role tele- communications has with regard to elec- tronic payment in shops (point-of-sale Figure 7 Interaction with content provider
8 Telektronikk 3/4.1998 are potential consumers of the content. other. The best example is mobile tele- that there is really little economic advan- In Figure 7 this picture is taken further phony which for a long time has been tage in doing so. Another aspect is that because the source of the content may supplementary to telephony, and has electronic media are not mature and are not be the service provider but some been a major reason for the general changing so rapidly that alliances which other company such as Levi’s or Coca increase in telephone traffic. However, are good one day may be bad the next Cola providing electronic commerce and with the penetration of mobile tele- day. C4 convergence may become advertisement, MIT offering distance phony we now have in Norway, it is important when this part of the business learning, Naxos offering music, or The starting to cannibalise the market for is stable. A survey of the opportunities Times offering an electronic newspaper. fixed network telephony. and the expected size of the markets are The service provider then offers storing found in [13]. of information and processing of support The point illustrated here is also related activities such as payment, security, elec- to C4 convergence. The four Cs stand for 6 Complexity and Orga- tronic marketplace, integration of con- content, computer, communications and tents and packaging. Telecommunica- consumer electronics representing each nisation: Co-operate tions as a part of these products is hardly of the four industries required in order to or Compete? seen by many of these content providers bring information to the user over elec- except as a cost factor, and possibly in- tronic media. They may form an hier- Above, we looked at the complete chain directly in terms of the efficiency of the archy or production ‘chain’ as shown in of activities from creation of content to service provider. the left part of Figure 8. Each industry displaying it to the user. In this chapter will also perform given tasks as shown we will look more closely at the commu- This new market is not mature; it is de- in the right part of the figure. nication part of the C4 chain, showing veloping now, and nobody knows how that it alone consists of a large number it will develop and when it is possible to The chain has consisted of independent of actors or activities. earn money from it (see for example [11] industries delivering their specialised and [12]). The problem is associated with products. As content is becoming more We have been used to regarding telecom- the duality of development time de- and more important, the interaction be- munications as one homogeneous and scribed above (see Figure 2). Most of tween the four industries is increasing. integrated business. This picture was in the required functionality is in place on They are also becoming more dependent many ways true until the monopolies in the Web or other commercial platforms on one another with regard to type of telecommunications were resolved: the for ‘fast lane’ development. However, we product, interactivity between consumer monopolies provided, as one single com- still lack some functionality such as con- and provider, quality of delivery and size pany, all types of telecommunications to necting these platforms with cash re- of consumer market. This has led to new everyone. In data transmission and gisters, databases, sensors and local pro- types of alliances, buy-outs and co-oper- mobile communications the disintegra- cessors, and to integrate them with sys- ation along the chain in Figure 8. This tion of the system into several roles has tems for logistics, transaction support is a game of gaining advantages: been emerging during the last twenty and security. This requires development • The content provider will ensure years. However, these businesses have in the ‘slow lane’ in order to bring for- access to good communication previously been regarded as being too ward these new basic technologies. channels and avoid competitors small to be representative of the evolu- gaining access to the same channels tion of the business. Electronic distribu- The markets for traditional telecommuni- and the customers connected to them; tion of content and the explosive growth cations are probably stagnating soon. For of the Internet are other tendencies which telecommunications operators it has been • The manufacturer of consumer elec- have caused the emergence of new roles very difficult to assess whether the mar- tronics may join with a content pro- ket is increasing, is stagnated, or is de- vider and make equipment dedicated to creasing. The reason is that it is difficult displaying the content for that provider to distinguish between different types of in order to increase his own market; traffic in the traffic machine: telephony, • Computer manufacturers may provide mobile communications and Internet. For dedicated equipment for processing Business Activity telephony it is also difficult to distinguish certain types of content and join with between voice traffic, telefax traffic and a communication provider in order to other data traffic. There are two other Content Create obtain a sizeable market for that prod- factors adding to this difficulty: uct; • Traffic is moving from traditional tele- • Communication providers may buy Process phony to the Internet because the latter Computer content providers in order to get ex- Store offer better functionality. The best clusive rights to distribute the content; example of this is telefax traffic which is replaced by e-mail. E-mail also • and so on. Communication Distribute replaces other telephone traffic because it is more convenient to write Several attempts on implementing C4 messages rather than speaking to an convergence in this way have been Consumer answering machine; made, but in general they have not been electronics Display successful. Either this type of business • The other reason is cannibalisation; integration is too difficult or it turns out that is, one service is replaced by an- Figure 8 C4 convergence
Telektronikk 3/4.1998 9 The roles are as follows: Subscription manager • Infrastructure owners own buildings, ducts, pipes, masts and other basic infrastructure required for accommo- dating cables and equipment. Most Clearing house existing telecommunications operators are in this role but it also contains Content host Retailer other parties such as road authorities, gas works, railways, cities and elec- tricity companies. One important Contract manager aspect of many infrastructure owners is their right-of-way which enables them to put up supporting infrastruc- Content provider ture efficiently. • Transport providers own cables, fibres, frequency licenses, satellites Platform operator System integrator and so on, and offer bandwidth, quality of service, capacity, multiplexing of Access provider signals and synchronisation of net- works. • Access providers offer interconnection Transport provider of users of telecommunications and networks. They own technologies like Infrastructure owner copper access lines, fibres, concentra- tors, frequency licenses, base stations, satellites, satellite earth stations, cable TV networks and so on. In Chapter 8 Figure 9 Business system we will look in more detail at the immense complexity associated with the access network and how different techniques are changing from being supplementary to becoming replace- and new businesses in telecommunica- Another important aspect determining able. This will serve as a good ex- tions. Still these new businesses are small the composition of the model is based on ample of the increasing complexity as compared to the traditional businesses the observation that for each role there of the system. of telecommunications companies. are in fact companies doing separate • Platform operators own switches, businesses there. Note also that the roles servers, processors and ‘intelligent’ In a model developed at Telenor the are atomic in the sense that it is difficult nodes for supporting processing, stor- telecommunications business is divided to divide them further without getting too ing, switching, interconnection and into ten roles plus the role of content pro- specific and losing the genericness of the routing of calls and content. Note that vider. The content provider role, as illus- model. a platform may be any type of process- trated in Figure 7, is in itself very com- ing device. It may be localised or plex and consists of a large number of The TINA Consortium7 has developed distributed. What is to be understood different and incomparable businesses. an equivalent model [15] for telecommu- by a platform is not how it is designed We have included the role in our model nications. Their choice of roles is purely or what it is called, but what it does. for completeness. The different roles are technical and its main focus is to deter- The platform may be public or private. interconnected with one another giving a mine which technical interfaces need to The platform operators are using model as shown in Figure 9. The model be defined. Our model is more focused equipment obtained from the transport in the figure is not complete for two rea- on the commercial issues and therefore providers for interconnecting the ma- sons: it shows only some of the possible this model contains a richer structure chines making up the platform and interactions between roles, and it con- than the model of the TINA Consortium. technologies of the access providers tains only one instance of each role. Sev- for connecting the users to their plat- eral roles of the same type may be pre- 7 Telecommunications Information Net- forms. sent in the same company, for example working Architecture Consortium serving different market segments or ser- • Subscription managers keep customer (established 1993) is a co-operation vices (one instance of the role serving the relationships, are responsible for bill- between telecommunications opera- residential market and one instance serv- ing, offer customer care, and interface tors, telecommunications manufac- ing the business market, or one instance the market. The customers may be res- turers and computer and software serving the market for fixed network ser- idential customers or businesses. manufacturers with the aim of speci- vices and another instance serving the fying the next generation distributed • Retailers sell subscriptions, traffic and market for mobile services). platform for telecommunications customer care to the market and may applications [14], [23].
10 Telektronikk 3/4.1998 be a market interface representing one and access operator (if they realise the or several subscription managers. Teledisc system). Bank • Content hosts offer storage and pro- There will be four generic interactions cessing to content providers. Depositing Loaning with the market. These are emanating • Clearinghouses offer transaction sup- from the subscription manager role, the Market port, security, information filtering, retailer role, the content provider role and trust and insurance against fraud and the system integrator role. These interac- Figure 10 Value network: banking liabilities. tions may also be simultaneous and have different content in different situations. • Content providers create and bundle The content of these interactions may be: information and may be any type of company or business offering content • The subscription manager is respon- such as newspapers, universities, mer- sible for marketing the overall product, of the goods from raw material to fin- chants, public offices and physicians. doing customer care and billing, and ished product, and logistics out repre- being the point of contact between the • Contract managers are in many ways senting how the product reaches the mar- customers and the rather complex equivalent to the subscription man- ket. The chain also contains several com- structure producing the services; agers. However, their main asset is to mon activities. It is, of course, more to it manage the contract between the con- • The retailer sells terminal products than this, but this would suffice in order tent providers and the providers offer- (television sets, PCs, telephones etc.) to understand the difference between the ing services to them. This will involve and manages the sales of subscription different value creation mechanisms. payment and billing services. contracts; Competition between value chains will be the traditional one for market shares. • System integrators design solutions for • The content provider may have any customers and bundle technologies and type of interaction with the market. The value shop is a problem solving contents from other actors of the This issue is too complex to consider organisations (consultants, architects and model. These are roles which can be here, and the content of this interaction physicians are examples) where value is filled by consultant companies, com- is not important for understanding created by solving problems for clients. puter industries and even advanced telecommunications. As will be dis- In telecommunications the system inte- content providers. The system inte- cussed below, the important point is grator is obviously a shop in this mean- grators do not need deep technological that it exists at the same time as the ing. It only takes place in the production skills or component expertise. They other interactions and may be indepen- process when building an advanced solu- need the skills of the entrepreneur and dent of them; tion. After that the system integrator should possess competencies like con- • The system integrator may have con- withdraws from the process. The man- sultative sales, knowledge of the value tacts with the market when co-ordinat- agement of the continuous process is production system of the customer, ing the establishment of a system or taken over by other roles. overview of available technologies connecting a new customer to it. This which may be used in the solution situation is common in the business The value network is represented by and their relative merits, insight in the market where the initial sale of a solu- organisations acting as mediating agen- expected evolution of the technology tion is undertaken by a system integra- cies. They may connect users of the same and the market, and ability to develop tor. Another more continuous contact kind as in the banking example in Figure new solutions. with the market may be to own the 10. The bank connects people with brand name for certain products. money to deposit and people wanting to The first six roles are those of traditional borrow money. In this case the network telecommunications. The other roles are One strategic challenge in telecommuni- does not even possess the required goods, new and have existed for about five years cations is to identify which type of namely the money. The goods are in fact or less, and are still in their embryonic organisation each role represents and owned by the customers. The network form. how value creation takes place in the may also consist of two types of co-oper- organisation. In [8] the following three ating banks: one having customers de- Note that manufacturers of equipment are value creation systems are considered: positing money, and one having cus- not included in the model. The reason is tomers borrowing money. In this ex- that the model contains only roles which • The value chain, ample the value of the customers of may take direct part in producing the • The value shop, and the co-operating bank is evident. This telecommunications product and have it extreme example shows one important delivered to the market. In this respect, • The value network. aspects of networks: the gain of co-oper- the equipment manufacturer is not visible ation. In this case the gain is obvious. In and plays no role in the production of These value creation system can be de- more complex cases it is not so obvious, services. Of course, a manufacturer of scribed as follows. The value chain is the and it may be difficult to assess the rela- equipment of software may participate in classical production system studied by tive merits of competition and co-opera- producing telecommunications services Porter [16] and is the type of value cre- tion. It may even happen that two compa- but, if so, in one of the roles defined ation which exists in the production indu- nies in network configuration may com- above. Examples are IBM as system stry. Simply spoken, it consists of a chain pete on certain products and co-operate integrator and Microsoft as platform of activities: logistics in representing the on others. There are many indications way raw material is obtained, production that this may be so in telecommunica-
Telektronikk 3/4.1998 11 Activity Content Advertisement obtaining Billing Create factory visibility Interaction Value shop Merchant Processing Process Transactions Newspaper Store Value network Delivery buying Distribution Interactivity reading Distribute Contact Customers Readers
Display Figure 11 Combined network and shop: newspaper Figure 12 C4 chain as value network
tions. Telenor and Telia co-operate in acts as a network connecting merchants per is primarily a value network but con- connecting customers in Norway and and their customers where some of the tains a value shop producing the adver- Sweden, while they compete in each customers of the merchant are also cus- tisements. The latter function could, of other’s primary markets. tomers of the newspaper. The merchant course, be undertaken by a separate is also a customer of the newspaper by agency. In this case the advertisement The network may also connect different buying advertisements. The figure illu- agency would have acted as a combined types of customers. Figure 11 shows a strates that different value creation network and shop. more complex case where a newspaper organisations may interact. The newspa-
System integrator
Creation
Processing Content Providers Contract manager
Content host Clearing Display house Subscription manager Retailer
Distribution Retailer
Subscription manager Platform operator Access operator Transport operator Infrastructure owner Figure 13 Business system and market interaction
12 Telektronikk 3/4.1998 Figure 12 shows the C4 chain as an inter- lines). The initiative for installing new Type of acting set of value networks representing equipment is taken by the system inte- Role organisation content, processing, distribution and dis- grator and the actual installation is done play. The main element here is that the by an installation company. Both may be Infrastructure owner Network providers of content, processing and dis- regarded as shops from a value creating Transport operator Network tribution will deliver their partial prod- point of view. The system integrator will Access operator Network ucts in parallel to the market. These also choose which companies should Platform operator Network deliveries are simultaneous and different, deliver the different elements of the sys- and none of them can be removed with- tem. This is the first phase of the delivery Subscription manager Network out collapsing the whole structure. Such of the product. In the next phase the Retailer Chain or shop a structure requires co-operation between installation company and the manufac- Content provider Any companies in different layers. Companies turer of equipment will not take part in Content host Network in one layer may co-operate with compa- the delivery. The product is then de- Clearinghouse Network nies which are competitors in another livered in parallel and simultaneously layer. A company distributing television by the energy producer, a company han- Contract manager Network signals may, for example, have as cus- dling the overall customer care and a System integrator Shop tomers different production companies company supporting the remote control. competing for viewers. The system integrator may not participate Figure 14 Classification of roles and main at all in this phase. We have included it value creating mechanisms The business model for telecommunica- in order to indicate that it may deliver tions is reproduced in Figure 13. Now it trust and guarantee in terms of brand is put on top of the C4 model in order to name and other ‘soft’ products. identify in which part of the chain each role is located. The figure also illustrates that the different parts of the model have 7 Complexity and • In the service area where the aim is to simultaneous market interactions. This Competition provide services over existing trans- applies not only to the ‘C4 interactions’ port and access structures indepen- but also to interactions between the mar- Competition between the incumbent and dently of the underlying network ket and individual roles of the business newcomers in the new liberalised market (point-of-sale, Internet); model for telecommunications. This is takes place in several areas: illustrated in a few cases: content host • In entering into network roles like • In the marketplace where the aim is to and subscription manager in the process- access, transport and platforms. get as many customers as possible; ing domain, and the subscription man- ager and the access operator in the distri- • In the access area where the main issue Competition in the marketplace takes bution domain. Note also that some roles is to install new accesses or to get hold place on price, on superiority of customer are present in both the C4 activities dis- of existing ones (leasing or local loop care, on products for special market seg- tribution and processing. The system unbundling) in order to bind the cus- ments both in the residential market and integrator has not been included in the tomers; the business market, and on bundling of C4 system. The reason is that one of the most important tasks of the system inte- grator is to create the complex value net- work required for each product. After this has been established, the system inte- grator is no longer a part of the business system delivering the product to the con- sumer. The type of value creating organi- sation of each role is shown in Figure 14. Energy producer • comfort service A final example is shown in Figure 15 • economisation also illustrating the interaction and co- chain billing operation between different value creat- Customer care ing organisations. The products are com- fort and economisation delivered by an Manufacturers network energy producer, where the product will remote • ovens control ensure the most economic use of energy • sensors Telecom for heating. The system consists of two installation chains, two shops and two networks chain network interacting in a parallel delivery of the Installation different components of the product. initiation Two phases of the product are indicated shop System integrator because different companies are involved brand in the delivery in the two phases: the shop name installation of the system (dotted lines) and the operation of the system (solid Figure 15 Interaction of shops and networks
Telektronikk 3/4.1998 13 where the aim is to capture parts of the of which may be poorly understood and High Mobile market already held by the other opera- involve undesired possibilities for arbi- telephony tor. The other competitor must therefore trage. Some of these aspects may be create a bigger market. This can only be uncovered by use of the method. Telephony done by increasing it towards the end of the market where mobility is not re- The second aspect is the importance of Rating garded to be very important but of inter- owning the access. In the GSM system est if the prices are sufficiently low. the operator owns the access in terms of These market segments are then ignorant frequency licenses. The user is owned by to the improved functionality provided not allowing him access to the competi- by mobility. In order to increase the tor’s network by technical means, at least Low Internet market in this way, the prices have to be in areas where the networks overlap. The lowered. This is an example where the subscription contract may also be such market in terms of turnover is increasing that if the user changes operator within a Quality FunctionalityFunctionality Price slowly – and in extreme cases may certain time, he will have to forfeit some become smaller, while the number of advantages. This situation is likely to Figure 16 Parameterised competition customers is increasing rapidly. The ben- change when the DCS1800 system is efit for the user segments requiring high introduced since the roaming restrictions functionality is then a lower price as dic- then are likely to disappear. The user tated by the new segment, while the ben- needs a primary contact with an operator efit for the market requiring the lower in order to manage his user profile. This price is more functionality: the mobile management and the actual usage of the telephone may be used to convey such system will then be two different activi- important messages as “I am now in the ties. This is again indicating the impor- train, and what are we having for din- tance of the business system and the products. Today much of this competi- ner?”. In the extreme case the prices for impact of the network economy descri- tion is concerned with offering telephony mobile communication and fixed net- bed in Chapter 6. There will then be two services in the residential market. In this work communications will become different concerns binding a customer as case, price will be the most important equal. In that case the two services have a subscriber of user profile and making competition instrument since there is moved from initially being complemen- him use a specific access and platform to little room for advantages on other prod- tary to becoming fully substitutional: perform the telecommunications. uct parameters. In accordance with [2] mobile systems extend the advantage of this is mainly a price war type of compe- the cordless phone without geographic In fixed networks the access lines require tition. Provision of the Internet as an limits. the major part of the investments in net- additional service is an added strategy works. Most of this network today con- where the focus is also on functionality Such competition may thus be partly or sists of copper wire, and it would involve (or quality as this parameter is called in wholly supplementary, complementary a major expense for a competitor to [2]). This changes the competition strat- or substitutional, and may change over replace it with modern technologies like egy as we shall see. time. Figure 16 shows another example optical fibres even though this would where fixed network telephony, mobile give advantages in terms of better Competition between mobile telephony telephony and the Internet are compared functionality and a wider spectrum of and fixed network telephony is escalat- using the three parameters (technical) services. One alternative for a competitor ing. Mobile telephony has increased the quality, functionality (as perceived by the will then be to lease access lines from the overall market for telephone services by user), and price (the comparison is based incumbent or whoever owns them. An- making the service more available. It is on a method proposed in [17]). Even other alternative could be for the incum- also replacing some of the need for fixed though all three networks may offer tele- bent to offer the access lines to the com- network telephony. This is also a type of phony, the Internet is the clear winner on petitors on conditions set by the regula- competition where price is not the only functionality, and this is likely to be the tory bodies (local loop unbundling). The parameter. key competitive parameter. Mobile tele- problem associated with such a decision phony also scores on functionality but is outlined in the next chapter. Competition between GSM operators is loses on price. Fixed network telephony an excellent example of price war com- cannot meet competition from the Inter- The third possibility is that the competi- petition. Mobile services are special in net by reducing the price because it can- tor offers an overlay service to the cus- one sense: they increase the availability not match the functionality. On the other tomers independently of the infrastruc- of people on the move. This is an advan- hand, mobile services and the Internet ture carrying it. The user will then be a tage of functionality. Initially this advan- may become supplements in the future customer of both the service provider and tage of functionality was exploited and by combining their functional merits. the owner of the underlying infrastruc- the prices were correspondingly high. ture. This is the way many Internet oper- The GSM system is such that it is very The method outlined above may be a ators do it, and is in accordance with the difficult to gain much from functionality: valuable method for analysis of competi- co-operation strategy of value networks. all parties offering the service can offer tion between products and companies. As Today the subscriptions for services and the same functionality at approximately the market moves towards technological infrastructure may be independent. As the same cost. Therefore, much is not convergence and C4 convergence, there this market expands, different subscrip- gained from a competition point of view will be a number of such examples, some tion forms may emerge (see the discus-
14 Telektronikk 3/4.1998 sion in Chapter 5 and 6 above). Co-oper- sector specific regulation of telecom- For some of the technologies only one ation may be beneficial for all parties in munications. operator can be present at a time. On a this structure. single copper wire or fibre there can only • Some access technologies are becom- be one access operator present. In other ing substitutes for one another. Substi- The final case is where the competitor technologies there may be several opera- tution may come as a result of the enters any of the roles defined in Chapter tors offering the same system at the same development of a certain technology 6 in order to compete in single or com- time. Radio systems are examples of such as optical fibres substituting bined niches. Here again the strategic these. In both cases each operator needs copper wires, or because different questions are related to how value is cre- an access infrastructure such as cables technologies may offer similar service ated and the overall gain for all parties and base stations. However, the alterna- capabilities such as radio access sub- from an increased market caused by co- tive to the first is that two or more sys- stituting copper wires. operation. tems (wires or fibres), one for each com- • Some technologies are supplementary petitor, are built to each customer, then Competition – or co-operation – in the to one another; that is, one technology leaving the choice of operator entirely market is extremely complex. Above we adds capabilities to another technology with the user. This alternative is, of have only pointed out some of the prob- enabling enhanced offering of services. course, economically not feasible except lems. Each of them requires advanced In this way an Internet access may be a in a few cases (existing cable TV systems analysis tools like system dynamics, supplement to a copper wire access and copper wire telephony). optimisation and game theory (though it may be a substitute for the telephone service) enhancing its basic A final problem is that some accesses are 8 Complexity of the features. unique for each user (wires or fibres) while others (radio) are shared by many • Some technologies are complemen- users. This also gives different Access: An Example tary, that is, they offer capabilities economies to different access techniques. which neither replace nor enhance The above considerations are rather theo- In some of these systems the cost per each other. One example is that copper retical. Therefore, it will be useful to user may on average be the same inde- wire access and radio broadcasting are consider one example just to illustrate the pendently of the number of users (copper complementary access technologies. complex analysis which is required only wire or fibre). In other cases, the access in order to understand the basic elements cost is reduced as the number of cus- One problem is that the categories are making up the system. A more compre- tomers increases (mobile telephony), or seldom pure. Access types may, for hensive analysis will require several the access cost per user may increase if example, be substitutes for each other in methods: dynamic performance of the the number of users increases (in mobile some cases and complements in other sit- business system using cybernetic and systems where the cell structure must be uations. Copper wire access is comple- econometric methods, stochastic analysis rearranged). mentary to some cable TV systems but and determination of the impact of uncer- competitive to others. tainty [18], game theoretic analyses With all these possibilities, access be- determining the merits of co-operation comes a multidimensional problem Another problem which complicates the versus competition in certain cases, and which is very difficult, if at all possible, issue further is that evolution of tech- other methods such as analysis of compe- to attack by use of conventional means. nologies, markets and prices may change tencies [19]. Each role in the business The dimensions are at least the follow- the category to which a given set of model of Chapter 6 needs to be modelled ing: accesses belongs or even enable new separately. Here we will give an analysis technologies to be developed such as the • Replacement capability as discussed of the access role and identify which ‘electricity modem’ on the local electric- above; parameters that need to be put into the ity network. GSM and copper access more comprehensive cybernetic, stochas- • Access type segmentation; were initially complementary, that is, tic and game theoretical models. GSM was mainly used by customers for • Basic infrastructure; whom mobility was important and did Access operation is chosen because it is • Geographic segmentation; not affect their usage pattern of fixed net- one of the most complex roles of work telephony. The price level is now • Market segmentation; telecommunications. There are several becoming such that GSM replaces copper reasons for this: • Service segmentation; wire access offering more functionality. • The control over the access and the DECT8, UMTS9 and other mobile tech- • Evolution and innovation; ownership of the users are regarded to nologies may do the same. • Functionality and bundling; be synonyms leading to various strate- gies such as building separate access • National versus international access; infrastructures, leasing access lines, • Pricing method. use of prefixes and other discriminat- ing means in order to accommodate 8 DECT is the digital European cordless This gives us a ten-dimensional problem several operators on the same access telecommunications system. where, for example, unbundling must be infrastructure, bulk sales of access 9 UMTS is an abbreviation for Universal understood in all dimensions. The first capacity, and local loop unbundling Mobile Telecommunications System item – the replacement capability – was where the conditions for obtaining which in most respects can substitute discussed above. The others will be con- access infrastructure are part of the the ISDN. sidered briefly below. There may be
Telektronikk 3/4.1998 15 Copper
Fibre
MarketMarket segmentsegment All segments All
ReplacementInteraction Mobile
Substitute
Residential
Supplement
Complement TechnologyTechnology Business Copper Fibre Mobile Figure 17 Example of ‘product space’
several more parameters which should 8.1 Access Types • DECT has the same capabilities as have been included. GSM and may replace it. Or it may The basic access types are (omitting dis- offer local mobility and thus not re- A given system will occupy one point in appearing technologies like NMT): place GSM but take traffic away from this ten-dimensional space and follow that system and fixed networks. How- • Copper wires capable of analogue one trajectory as the system evolves in ever, it is more likely that the system transmission or ISDN. ADSL10 and the future. Another system will occupy a may be replaced by other mobile tech- similar technologies are supplements different point now and follow a different nologies since the concept is becoming offering more bandwidth. The basic trajectory into the future. Two systems obsolete before it is installed. technology exists everywhere and sev- will most likely not be identical and eral supplements are possible, but their • TETRA11 is primarily a system for occupy the same points of this space. market potential depends mainly on private trunked radio but can be used This makes it extremely difficult to iden- price. The bandwidth that can be used for public access and compete with the tify when rules defined for competition on wires is therefore increasing, which GSM system, at least in cities. The are met or violated. An example is shown may imply that existing cables have a technology is based on the GSM tech- in Figure 17. We have included only the much longer lifetime than expected. nology. three parameters technology, market seg- ment and interaction, and shown a pos- • Optical fibres offering ‘unlimited’ • DCS1800 is a GSM technology in the sible evolution at two instants. Here we bandwidth to the user. The technology 1800 MHz band offering more band- have assumed that copper wire access is exists in the business segment and is width. This technology is now being moving from being a supplement in all coming in the residential market as implemented in several countries, and markets to becoming a substitute (that is, cable TV networks are enhanced and the manufacturers are making dual one of many techniques) in all markets, copper wires are replaced in areas mode mobile stations operating both in optical fibre access is shifting from being where cheap infrastructure for laying the 900 MHz band and the 1800 MHz a substitute in the business market to also fibres exists. band, and several regulating authorities becoming a substitute in all markets, and are requiring roaming to be possible • GSM in the 900 MHz frequency range mobile access is moving from being a between systems in the two bands also allows roaming between almost all complement in the business market to when the operators compete in the countries, and is besides copper wire becoming a substitute in the residential same geographical area. the most used access technology in the market. The question is: when and where world. are these technologies competing? 11TETRA is an abbreviation for Trans- 10Asynchronous Digital Subscriber Line. European Trunked Radio Access.
16 Telektronikk 3/4.1998 • UMTS will probably replace GSM as • Internet over any of the above basic of frequency licenses will probably the next generation land mobile sys- technologies; become one of the most important tem, or at least this is the intention competitive assets in the future. • Modem access to data networks over behind specifying the technology. wire, radio or fibre; A new entrant can utilise a combination • Combinations of terrestrial radio solu- • Frame relay over wire, cable and fibre; of infrastructures in order to find the tions are possible. Examples are minimum cost for building a new system. GSM/DECT and GSM/DCS1800. • Multimedia access in the future over Local loop unbundling and frequency any of the above. • Cable TV networks offering one-way licensing are regulatory matters which broadcast services is also an access will have a major impact on the competi- Some technologies have been developed technology. This technology may also tion in the market. Local loop unbundling but are still not available because there become a combined substitute and is particularly important since it may has not been a big enough market for supplement for some applications on open for arbitrage (buying capacity and them. This may change radically as the the copper wire. The systems will have portioning it into smaller units, for ex- liberalised market is maturing and the the capability of offering broadband ample, dividing a 64 kbit/s channel into control of the access becomes an increas- Internet services to the user and com- four 16 kbit/s channels by use of voice ingly important issue for the newcomers. bining it with narrowband technology coding techniques), it may leave uncer- for the return channel in co-operation tainties with regard to responsibilities with another operator. This will reduce 8.2 Basic Infrastructure and liabilities, and it may slow the evolu- the time for downloading information tion of access technologies since it re- from the Internet considerably. The existence of a basic infrastructure moves the need for this type of innova- and the cost of building a new infrastruc- tions. • Cable TV networks can also be en- ture determine the threshold for new hanced for two-way services as has entrants to establish themselves in the been done in several countries. 8.3 Geographic Segmentation access market. The access infrastructure • One-way radio broadcast networks is the most expensive part of the network The residential market and the business (terrestrial or satellite) may be en- since the possibility of utilising traffic market are rather concentrated in many hanced in the same way as cable TV concentration and dividing the invest- countries, and a new entrant will enter networks. ment costs on many accesses is small. selectively in areas which will offer the Infrastructure elements which may be • Geostationary satellites can offer best business opportunities. There will used are: broadband transmission one-way and certainly also be areas where no-one will combine it with a narrowband system • Ducts, pipes and tubes like sewers and enter. The geographic segmentation will for the return direction, for example, gas pipes, or ducts deliberately made be of particular importance for the offering Internet services. for various use in the future (along incumbent because in most countries roads, railways and so on); dominance is defined in terms of market • Mobile geostationary satellite systems share (for example dominance may mean such as the INMARSAT system. • Poles for road lighting and electricity that the total market share is above supply suitable for mounting radio • Low orbit satellite systems such as 25 %). This may mean that a company antennae for mobile communications IRIDIUM and Teledisc. may hold most of the subscribers in a systems like TETRA and DCS1800, country and being dominant without • ‘Eelectricity modems’ enabling 1 or and for placing sensors like microchip earning money since the competitor has 2 Mbit/s over electricity supply sys- video cameras; taken the most lucrative niches offering tems in the residential market. • Electricity supply systems that can be good margins. • ‘Balloon satellites’ where cheap satel- used for roll out of optical fibres, in lite repeaters are placed in balloons for particular to the industry; Several values may be given to the geo- medium area coverage for both two- graphic segmentation variable: • Unbundling the copper line access of way systems and broadcast systems. the incumbent where a competitor is • The country is regarded as a single • Broadband two-way land mobile sys- buying or renting the existing systems; segment with regard to market share tems (BMS) are being developed offer- and competition and does not take into • Utilisation of the same duct for several ing capacity comparable to that of account geographic concentration of purposes when new areas are built out optical fibres. people and businesses; or some infrastructure is maintained or replaced in an area. This is particularly • A distinction is made between cities Note that most of these technologies important for electricity companies if and rural areas in the rules governing were put into service or developed during they also offer telecommunications market shares; the last five years, and that some of them where they may distribute the cost on are still being developed. Note also that • The country is divided into areas such two (or more) products (energy and in many cases two accesses are required as counties or municipalities and the telecommunications); in order to establish the pertinent connec- market share of competitors may be tion between the user and the service • Radio frequencies, where in some computed for each segment of the grid. provider. Examples are cases the obstacle is licensing of fre- quency usage making this to a force It may then be possible to put different which is not entirely market driven. weight on the segments and competition On the other hand, the availability within each segment:
Telektronikk 3/4.1998 17 • Each competitor should have the same tion will be typical for premium rate ser- Noting that the applications in terms of capability as the incumbent in every vices and for Internet services. Travel telecommunications services as we are segment of the grid; agencies and taxis are examples of such used to them are not offered by the asymmetric customers. access itself but by the platform to which • All competitors combined should have the access is connecting the customer, the same capability as the incumbent, Competitors will search for market seg- this variable may take on a large number but there may be different competitors ments offering much income and healthy of different values. The access may then in different geographical segments; margins, and ignore others. Therefore, offer capabilities as follows: • There may be areas where there are direct competition may not occur in cer- • Narrow bandwidth offered to residen- natural monopolies. These areas are tain market segments, while other seg- tial users with standard supplementary not considered with regard to market ments will have many competitors. The features; share; market in one way resembles the car market where there is no competition • Narrow bandwidth offered to busi- • Only the market share for the core between Skoda and Mercedes because nesses; businesses of telecommunications they attack different market segments, (telephony, transport systems and • ISDN (two channels and more flexi- while there are other manufacturers copper wire access) is included in the bility) offered to residential users; offering products to several segments evaluation but not other technologies. competing with both of them (for ex- • Extended ISDN offered to businesses; In such regulations it is, for example, ample the Ford company and most ignored that telephony over copper • Broad bandwidth offered to businesses Japanese companies). The competition is wires is competing with mobile sys- and residential users; so new in telecommunications that this tems; type of business has not yet developed. • Dark fibre offered to businesses; • It is recognised that different techno- • Asymmetric solutions where there is a logies compete, and the definition of The market segment variable may then broadband system from the network to competition and market share is be characterised by the following vari- the user, while the opposite direction is widened to take this into account. ables: served by a narrowband system; • Residential market with several sub- The different strategies will define differ- • Supplementing (and possibly more segments where each segment is ent competition conditions, and it may expensive) technologies like ADSL. characterised by usage volume and lead to different strategies for co-opera- usage patterns; tion between the various actors in the 8.6 Evolution and Innovation marketplace. • Business market divided into cate- gories based on the size and type of the The competitive environment must be business and with several subsegments 8.4 Market Segmentation such that it favours evolution and innova- based on traffic direction such as high tion. This is probably the most difficult outgoing traffic, high incoming traffic, The telecommunications market is com- issue to regulate for several reasons: or both. plex and consists of several segments. • Unbundling of the copper wire may First, we have the important differentia- give all competitors equal access to the tion between the residential market and 8.5 Service Segmentation users so that there is no competitive the business market. Then within each of force favouring evolution of alterna- these there are several segments depend- Note that here we refer only to services tive and better solutions. Under such ing on total usage, usage patterns and offered by the access and not to services regulations, will fibre to the home service requirements. offered by a platform to which the access come at all? is connected. These cases must be kept Within the business market one impor- distinct because the access deals only • How should the operator in such a tant factor is the size of the customer and with the way in which value added ser- regime finance a joint evolution of the the amount of expenses the company vices are offered and not how they are technology? uses on telecommunications. Another produced. I mentioned above that the • How should an operator enhance the not so obvious differentiating factor is Internet will require an additional access network if the use of it is sold to an- whether the customer produces much on top of the access offered by copper other operator? Who should pay for outgoing traffic or much incoming wires, fibres and radio. In this case one such innovation or will there be any traffic. It is easy to measure the former access technology is concerned with how sharing of risks? Who is responsible and determine how much the operator the user is connected to the network. The for faults? earns from that user. However, the sec- other access (to Internet) ensures that ond type of customer may create much the interactivity between the user and • Access from individual customers are income from the users calling this cus- the network is maintained. often multiplexed on the same trans- tomer or from interconnect fees for ter- mission system in order to reduce the minating traffic from other operators. In Some technologies may only offer a cost per access. This increases the the business structure described above limited set of the access services listed problems of responsibility and liability (Figure 9), the income from traffic may below. Other technologies may offer a if something goes wrong. thus come in one role while there is an- larger or more sophisticated range of other role being responsible for the cus- services. This problem is associated with the prob- tomer terminating the traffic. This situa- lem of defining the extent of the access
18 Telektronikk 3/4.1998 network, also noting that this is technol- A less visible bundling comes from oper- 9 Conclusions ogy dependent and subject to evolution. ators offering Internet telephony. This The fixed network has evolved towards product is poorer with regard to quality The most important strategic competency fewer exchanges (10 % of what we had and reliability than ordinary telephony, for companies within telecommunica- 20 years ago) but with a large number of but what makes it attractive is not only, tions is to understand the complexity concentrators and remotely controlled as is the general feeling, that it may be of the business and to act in accordance switching units with limited service capa- cheaper but that it offers more function- with this understanding. The source of bilities. This evolution may continue ality being associated with text and gra- much of this complexity is new, and the depending on the general evolution of phical services. The service may be business is changing for many reasons: transmission systems. On the other hand, triggered by clicking hypertext pointers • The telecommunications technology a land mobile system contains fewer in the same way as for accessing the has been through a rapid change dur- switches and has a much simpler trans- Internet from locally stored information. ing the last decade. Not only has the mission structure. It requires a large This service is offered from encyclo- focus changed from the primary con- number of base stations so that the access paedia on CD-ROM for access to supple- cern of being competent in transmis- is not necessarily cheaper. This makes it mentary information or updates: buying sion to putting content in the front seat, difficult to compare them and to see the CD-ROM, you also buy the access to but also to identify and master applica- where the evolution will take them with information. tions which can be made from general regard to market penetration and compe- purpose processing. This part of the tition in the future. All of the technologies listed above evolution is still going on but a new should be characterised with regard to component is entering: technology It is also reasonable to assume that it is their differentiating capabilities, both now moves towards cheap sensors and more likely that new access technologies alone and in different bundles. nanotechnologies opening for new will be developed if the competitors are applications in monitoring the environ- forced into a situation where they have to 8.8 National versus ment, people and machines; be innovative rather than copying old and International Access obsolete technologies. This may give the • The telecommunications industry incumbent some initial advantage but developed the transmission technology We always assume that competition is technological evolution may take that but has not been the driving force in national and do our assessment based on away soon (the ‘electricity modem’ may the evolution which has taken place that. However, low orbit satellite systems be such a leap). afterwards. The data industry, the con- (IRIDIUM, Teledisc) may be important tent industry and the academia have threats to national access operators. Such been the main driving forces in this 8.7 Functionality and Bundling systems will produce relatively little development. The microelectronics delay in the speech channels so that they industry may enter next in the develop- Market penetration is not only dependent may offer a quality of service not signifi- ment of the new sensor technology; on the cost of the access. This is the case cantly different from that of the fixed if the product is a primary product. As network. • The dependability of society on we defined it above, telephony in the res- telecommunications is increasing. This idential market is such a product. We Other types of competition will be met applies to all parts of society and to also saw above that it is possible to bun- from other satellite operators. Internet more and more of its activities, and dle it with other products such as cable over broadcast satellites is again a good the new technologies will make this TV, energy and the Internet. This may example. This may change the telecom- dependability even more critical. change the value of the telephony prod- munications business completely because Dependability and telecommunications uct as perceived by the user. It may also geographically unbound operators may should become a new area of academic change the price of the product, either by take over part of the business. This is not studies; cross subsidising or by lower cost, unlike the way in which long distance • The organisation of the business is also because some cost elements can be dis- satellite communications removed the changing rapidly. This is partly be- tributed over more products in the bun- lucrative businesses of transit traffic. cause of liberalisation of the market, dle. The intelligent building is such a but the technological evolution has concept where the cost is divided among 8.9 Pricing Method probably been just as important for products like telephony, energy delivery this change. The new technologies and surveillance. In this case, it may be The access competition may be obscured will bring this organisational change argued that the bundling is entirely because of different pricing mechanisms. even further. Telecommunications within telecommunications since the user We have already seen one such problem may then become a commodity pays for the remote control and monitor- related to bundling where cost is divided supporting other applications and be ing facilities. among several products. Other possibili- integrated in other businesses leading ties are: to an advanced form of C4 conver- This part of the problem is very complex gence. Making money in such an en- and the possibility and capability of • Access financed by advertisements; vironment depends very much on bundling may change over time and • Access financed as part of house understanding how value is created hence become an evolutionary parameter. rental; in the organisation [8]; • Access financed by value added ser- • The products are also taking on new vice such as electronic commerce. forms, changing from being mainly
Telektronikk 3/4.1998 19 two-party services to accessing content References 11 Hagel III, J, Armstrong, A G. and information. This evolution is in net.gain. Harvard Business School many ways a result of the growth of 1 Dendrinos, D S, Sonis, M. Chaos and Press, 1997. the World Wide Web and its support- socio-spatial dynamics. Springer- ing technologies. The advance of the Verlag, 1990. 12 Cairncross, F. The death of distance. sensor technology may again change Harvard Business School Press, this picture, and that even before the 2 D’Aveni, R A. Hypercompetitive 1997. previous technologies and businesses Rivalries. The Free Press, 1995. have matured and stabilised; 13 Baldwin, T F, McVoy, D S, Stein- 3 Saffo, P. Closing session of ITS 98, feld, C. Convergence : Integrating • This indicates that mastering advanced Stockholm 21 – 24 June, 1998. media, information & communica- methods in strategic planning may tion. Sage Publications, 1996. become key factors in the future posi- 4 Gemperlein, J. Paul Saffo, tech fore- tioning in the market. These methods caster. San Jose Mercury News. 14 Handegård, T, Kristiansen, L. The should build on game theory, cyber- Available on World Wide Web under TINA architecture. Telektronikk, 94, netics (both feed back and feed for- the title Q and A with Paul Saffo. (1), 95–106, 1998. ward methods), optimisation and spatiotemporal evolution theory ([1], 5 Regis, E. Nano! Bantam Books, 15 TINA Consortium. TINA Business [20], [21]). 1997. Model and Reference Points, Version 4.0, May 20, 1997. There is no other industry where similar 6 Cochrane, P, Heatley, D J T. Mod- changes have occurred, so that there is elling future telecommunications sys- 16 Porter, M. Competitive advantage : not much to learn from experience and tems. Chapman & Hall, 1996. creating and sustaining superior per- analyses of other businesses. The strat- formance. Free Press, 1985. egy for telecommunications must there- 7 Audestad, J A. Challenges in a liber- fore be built from scratch. alised telecommunications market. 17 Kim, W C, Mauborgne, R. Value OR98 Conference Proceedings, innovation : creating and sustaining Most of the development of strategy as a Zürich, August/September, 1998. superior performance. Harvard Busi- scientific subject has taken place since ness Review, Jan.-Feb., 1997. 1970. The full recognition of the subject 8 Stabell, C B, Fjeldstad, Ø D. Con- came around 1980 when the Strategic figuring value for competitive ad- 18 Wallace, S W. The role of uncer- Management Journal was founded. Still vantage : on chains, shops, and net- tainty in strategic planning. Telek- strategy is an issue which is not fully works. Strategic Management Jour- tronikk, 94, (3/4), 1998, 21–23. (This recognised. The best methods are still not nal, 19, 413–437, 1998. issue.) in common use, they are simply too com- plex. In telecommunications we probably 9 ISO/IEC/ITU. Reference Model of 19 Audestad, J A, Kjæreng, A, Mahieu, need them. One mind opening survey is Open Distributed Systems Part 2 : L. An object oriented model of the contained in [22], showing how complex Descriptive Model. ISO Document telecommunications business. Telek- strategies may be. 10746-2/ITU-T Recommendation X. tronikk, 94, (3/4), 1998, 54–61. (This 902. issue.)
10 Audestad, J A. Strategy for enterprise 20 Rumelt, R P, Schendel, D E, Treece, specifications. Telektronikk, 94, D J. (eds.). Fundamental issues in (3/4), 1998, 24–32. (This issue.) strategy. Harvard Business School Press, 1994.
21 Anderson, P W, Arrow, K J, Pines, D. Economy as an evolving complex system. Santa Fe Institute Studies in the Science of Complexity, Volume V, Addison-Wesley, 1988.
22 Kelly, K. Out of control : the rise of neo-biological civilization. Addison- Jan A. Audestad (56) is Senior Advisor for the Corporate Management of Telenor. He is also Adjunct Professor of Wesly, 1994. telematics at the Norwegian University of Science and Technology. 23 Korzeniowski, P. Middleware : achieving open systems for the e-mail: [email protected] enterprise. Computer Technology Research Corp., 1997.
20 Telektronikk 3/4.1998 The role of uncertainty in strategic planning
STEIN W. WALLACE
1 Introduction market division of the company is asked But then a stubborn member of the staff how many customers the company can insists on checking also the expected per- Strategic planning and uncertainty1 are expect, the reply is that they expect formance of the mobile solution, despite twins. The first makes no sense without 6,000, but, in fact, they could get any the fact that no possible customer base the latter. It is the very existence of number of customers between zero and exists where a mobile solution is best. uncertainty that forces us to produce 10,000. This would depend on their own And, lo and behold, its expected perfor- strategic plans. Had we known the future, strategies, as well as those of the com- mance is substantially better. What is strategy would have been without inter- petitors. However hard pressed, the mar- going on? Figure 2 illustrates the point. est, even without meaning. ket division refuses to be more specific While the mobile solution is never best than this. The management therefore when the number of customers is known, This very obvious observation leads to turns to its planning division with this it is at times much better than any of the some conclusions that in my view ought information and asks what would be the fixed solutions whenever the number of to be as obvious. If uncertainty is an correct decision in the different cases. customers is different from the best pos- important part of the problem, it is rea- The planning division replies by pro- sible value. sonable to think that it must also be an ducing a figure, as the one in Figure 1. important part of the methodology used 3 What have we learned? in planning. By methodology I now refer From Figure 1, management sees that to both quantitative models and qualita- whatever number of customers it obtains, There are many things to be learned from tive reasoning. And even more impor- it will be optimal with a fixed network. such a simple example. Because, indeed, tantly, it is not at all unimportant how the After all, there exists no possible future the example is very simple. So simple, uncertainty is entered into the planning (i.e. customer base) where mobile is that it is tempting to think that the logical process. better than fixed. It is therefore decided errors made above are unreasonable. that they will build an infrastructure for a Also, it may be tempting to start dis- Before continuing, let us note in passing fixed network. A further analysis is per- cussing the model rather than the con- that although we are now discussing formed to determine the size of the fixed clusions, based on the feeling that the strategic planning, most, if not all, of our installation. The approach is as follows. problems are caused by the simplicity conclusions also apply to tactic planning. Each of the ten solutions obtained to pro- of the model. That, however, is not the In my view there are no deterministic duce the columns for fixed networks in planning problems. There are certainly Figure 1 is tested towards all possible planning problems that are well studied futures to find which one has the best by deterministic methods, but how do we expected performance.3 For example, 3 Of course, we could also maximize know that the one we are looking at is in we check how the solution for 3,000 cus- expected utility, or even perform a that class? The rest of this note will indi- tomers behaves if in fact 7,000 customers worst case analysis. The arguments cate what may go wrong if we rely on show up. This way the best of the fixed to follow will function also for these deterministic tools in what is truly a network solutions is found. cases. stochastic problem.
2 A small example2
Assume that a provider of telephone services, operating under competition, is faced with the problem of how to develop its infrastructure in a residential area under development. The infrastruc- ture must be built while the residential area is under construction, and before Cost pr customer any customer has decided which tele- 1200 phone company to use. The area is Mobile planned for 10,000 housing units. In 1000 monopoly times this meant almost surely Fixed 10,000 customers. But now, under com- 800 petition, things are different. When the 600
1 In this short note we do not make any 400 distinctions between stochasticity, 200 uncertainty, randomness or risk. In some contexts that may be important; 0 here it is the relationship between 12345678910 stochastic and deterministic that is in focus. Customers (x 1000) 2 This example was created in co- Figure 1 Cost estimates from the planning division. For each possible customer base, operation with Nils-Jacob Berland. the cheapest mobile and the cheapest fixed solution is given
Telektronikk 3/4.1998 21 solutions is the right thing to do. After all, one of them will certainly be optimal after the fact. And still, we might be totally mislead. The problem is of course Cost pr customer that if we pick the wrong scenario solu- 1000 tion, it may turn out to be very bad. Mobile Fixed 800 3.2 Evaluating decision mak- ers 600 As we have seen, decisions which are 400 optimal in terms of expected behavior are normally never optimal in hindsight. Hence, decision-makers that follow our 200 advice, and which are the best decision- makers in the long run, will never make 0 the really outstanding decisions. This is 12345678910 important to remember when decision- Customers (x 1000) makers are evaluated. Do we value the decision-maker who occasionally makes Figure 2 Principal illustration of how a single solution behaves if the customer base marvelous decisions, or the ones who are is different from the one corresponding to that solution good in the long run? So care must be taken with respect to hindsight, as the IQ of hindsight is very high.
4 So what are scenarios?
We are of course free to define scenarios case. The erroneous thinking used above metric analysis to see if the first solution any way we want. But for scenarios to be is what we find in almost all textbooks, is stable. If he finds that the solution useful in a context of decision-making, and many software packages advertise remains optimal (or aspects of the solu- they must have one crucial property. that they facilitate exactly this type of tion remain intact) for many possible A scenario must be independent of the approach. parameters, he feels confident that the actions of the decision-maker. A scenario first solution was good. What we saw in therefore describes the (random) environ- What went wrong in our argument, and the example is that this is false. Not only ment facing the decision-maker. Hence, what is indeed a common error, is to base does the parametric analysis not yield statements such as “it will rain tomor- decisions on the following assumption. what it promises, it is directly counter- row”, “EU regulation will allow compe- If we find the optimal decision (plan, productive. Before the analysis the deci- tition” and “demand will increase” are strategy) for all possible futures, and all sion-maker was uncertain about the qual- most likely scenarios, whereas “we will these decisions share some property, we ity of the solution found from the deter- choose to cooperate with Telia” is not a feel confident that this property must be ministic model, afterwards he feels more scenario (unless the decision maker part of the solution we are seeking, confident, despite the fact that it is in- views decisions in his own company as namely the one which is best in an over- correct. The analysis has fooled him. We random events he cannot affect or con- all (expected) sense. We think that by see that sensitivity analysis simply is not trol.) studying scenario solutions we learn a way to understand and analyze something useful about the problem stochastic decision problems. A more de- Looking at our example, we may wonder under investigation. As we saw above, tailed discussion of this fact can be found if the number of customers is a good ran- that is not at all a generally correct ob- in Wallace [2]. dom variable, i.e. have we defined good servation. For absolutely all possible scenarios? If we can affect the number of futures the best was to install a fixed 3.1 Optimal in hindsight customers for example by marketing and network, and that was exactly what we price schemes, our scenarios are not did not want. particularly well defined, as they are not For the example, note that in hindsight, independent of our actions. Rather, that is, after the fact, one of the scenarios Often, the starting point of an analysis is we should have tried to define scenarios will have happened. Hence, one of the a deterministic model where all random in terms of general trends among cus- scenario solutions will be optimal. There- variables are replaced by their means (or tomers, governmental regulations, and fore, in hindsight, it will appear that a some other representative). The para- possibly strategies of competitors. Sce- fixed network solution is best. We know meters are often not explicitly interpreted narios may be defined in terms of such as for sure that the solution we advocated, as means, but implicitly that is what they the general popularity of mobile solu- namely a mobile solution, will not look are. This will produce some decision tions, whether or not our competitors extremely good. This is important to (strategy). Since the decision maker will develop a fixed network, and to what note, as this is probably a major reason knows that the parameters used may be extent we will be allowed to compete. for imagining that looking at scenario incorrect, he then resorts to some para- Decisions may then be such as market-
22 Telektronikk 3/4.1998 ˆ ing, market research, and, of course, our max ∑ p(s')f (s',xs ) to obtain x. 6 Conclusions x for s S choice between a mobile and a fixed net- {}s ∈ s'∈S work infrastructure. The main message of this short note has been that the use of scenarios for finding What we have found out earlier is that xˆ It is important to note what we argued potential solutions (strategies) for a de- can be a rather bad solution, despite against in the first section. It was not the cision problem under uncertainty (be it being the best of the scenario solutions. use of scenarios as such, nor was it the strategic or tactic) can lead to arbitrarily Instead we should solve fact that the scenarios possibly were not bad solutions. Scenario solutions do not too well defined. The point was that sce- offer information about the structure of max ∑ p(s) f (s, x) to obtain x* . narios cannot be used to find good solu- x s∈S (1) the solution that is best in expectation. tions, since good solutions rarely are Hence, other, and more complex methods solutions to scenario problems. But are needed, and these do indeed exist. The latter problem is of course much scenarios are appropriate for evaluating harder to solve, as it is much bigger. possible solutions. That is, given a solu- Scenarios are on the other hand appropri- It is also harder to grasp mentally if tion (strategy), and a set of scenarios ate for evaluating existing potential solu- one resorts to qualitative approaches. describing the possible futures, we can tions. Technically speaking, this is what Because also in the qualitative setting, a correctly find the expected performance we normally refer to as simulation. scenario based approach will lead in the of that solution. This is a process we 4 wrong direction. would often refer to as simulation. So References scenarios are indeed important for deci- It is not the purpose of this note to dis- sion making under uncertainty, as long cuss the issues of formulating and solv- 1 Kall, P, Wallace, S W. Stochastic as they are appropriately used. ing stochastic programs, such as (1), but programming. Chichester, Wiley, rather to point out that such complex for- 1994. 5 Scenario aggregation mulations may be necessary, and that procedures are available. For a detailed 2 Wallace, S W. Decision making Methodologies for using scenarios to find look at the field, see for example the text- under uncertainty : is sensitivity optimal solutions to planning problems book by Kall and Wallace [1]. analysis of any use? Operations Re- do exist. Having understood that the dif- search. To appear. ficulty with scenario analysis is its in- 5.1 Decision trees ability to find optimal solutions, and not its ability to evaluate potential solutions, Many are used to applying decision we realize that what we need is a more trees for studying decision-making under advanced setting for the optimization. uncertainty. A decision tree is a tree that (And this applies also to decision-makers has two types of nodes, one for possible who make decisions based on qualitative decisions, and one for realizations of analysis.) Verbally, what we need to do random variables. Decision trees are log- is to set up a model that searches for a ically correct, and yield optimal solu- solution that is optimal in the expected tions. Note however, that for such trees, sense, rather than being optimal in one there must be a small and finite number single scenario. of possible decisions. Hence, the poten- tial actions to be taken are input to the Let s be a scenario (possible future), p(s) model, and the output is just information the probability that s will occur, and S the about which one is best. Hence, decision set of all scenarios. Assume our problem trees do not help us find potential solu- is to maximize the expected profit tions, and in that sense suffer from partly E f(s,x) with respect to x. Here E means S S the same problems as scenario analysis. to take the expectation with respect to the But if the number of potential solutions is scenarios in S, and f(s,x) is the profit if indeed clearly finite, decisions trees are we make decision x and scenario s appropriate. happens. So the goal is to find the x which has the best expected perfor- mance. The scenario approach, on the other hand, is to solve for one s at a time Stein W. Wallace (42) has been Prof. of Operations Re- max f (s, x) to obtain xs . search at NTNU since 1990 after spending 7 years as x Senior Scientist at Chr. Michelsen Institute, Bergen, and 3 years at Haugesund Maritime College. Throughout, his Then, to find which of these solutions focus has been on theoretical and applied aspects of behave the best, we solve decision making under uncertainty, with applications in fisheries management, petroleum field development, portfolio management and telecommunications. 4 Although the term simulation is used e-mail: [email protected] with many different meanings, and is often not very well defined.
Telektronikk 3/4.1998 23 Strategy for Enterprise Specifications
JAN A. AUDESTAD
1 Why Open Distributed tions equipment. There are several con- All viewpoints except the enterprise cerns behind this development: viewpoint are well described in terms of Processing? concrete methodologies and languages. • Telecommunications and information The enterprise viewpoint is rather vague Telecommunication is a technological technologies are converging so that because enterprises are so different that area and the technology must be the basis there are several commercial and tech- it is hard to find a common denominator for the definition of the strategy for such nological advantages of developing a for them. However, it is possible to pro- systems. Telecommunication systems are common platform for such systems; duce a rather comprehensive enterprise also distributed systems by their very • The TINA platform should replace the model for telecommunication systems. nature. It is, however, new that distribu- intelligent network nodes of the exist- tion has been taken into account in com- ing networks with more modern equip- puter systems in general. To a great ment with better scalability perform- 2 General Enterprise extent such systems have been modelled ance and more evolutionary potential; without clear considerations on distribu- Requirements tion. During the last few years, however, • The TINA platform should convert the For ODP systems a specific formal lan- distributed processing for general com- telecommunications networks into sys- guage based on object orientation is de- puter applications has been developed tems with the same flexibility as the veloped for each viewpoint. This is easy jointly by the ISO, IEC and ITU1 ([1], World Wide Web while meeting the to do when defining information struc- [2]). The idea is that this specification new quality of service requirements tures and when describing how computa- should become the basis for all distri- of modern transport networks like the tional processes are co-operating in order buted computer systems. The first com- ATM; to perform a common task. ISO has mercial products are available now • Telecommunications and information attempted the same for the enterprise (CORBA2). technologies are converging in such a viewpoint but, in my opinion, this way that the whole production chain attempt has not been successful, the ob- The specification is called Open Dis- consisting of creation, processing and vious reason being that it is hardly pos- tributed Processing, and is abbreviated storing, distribution and displaying of sible to give a single definition of what ODP. We will use this abbreviation content is merging into one inseparable an enterprise is. throughout. The purpose of this essay process – see [3]). is to use these specifications in order to The main issues at the enterprise view- demonstrate how this specification can The most important achievement in ODP point are to identify: be used to formulate a strategy of is the decomposition of the specification telecommunications systems in general. • The purpose of the system to be mod- into five viewpoints. This is in itself a In particular, we will concentrate on the elled, designed and implemented; major strategic achievement because it methods used for formulating the re- divides the specification into five sepa- • How this system is located administra- quirement specification of a system, in rate and non-interfering parts called tively, technically or otherwise within ODP referred to as the enterprise view- viewpoints. Each viewpoint must be the enterprise; point specification. specified for every system in order to • How it (or the enterprise) interacts Note that every system can – and make the specification complete. The with the environment; should – be specified at the enterprise viewpoints are: viewpoint using the principles pre- • Policies that the system should support • The enterprise viewpoint specifying sented here. These principles are or satisfy, where the policies may be the overall requirements of the system; general and independent of whether defined by the enterprise itself or the system in question is an ODP sys- • The information viewpoint specifying forced upon it by the environment tem or another type of system. the information content in the system (for example a regulation policy); and how individual information ele- • Requirements concerning perform- Since 1993 work has been progressing on ments are managed; ance, costs, reliability, availability, a new architecture for telecommunication • The computational viewpoint specify- dependability, etc. based on this model. The work has been ing how the system behaves in order to a co-operation within the TINA3 Consor- complete the tasks allocated to it with- When an enterprise decides to implement tium owned jointly by telecommunica- out taking into account how this pro- or purchase a new system, say, a local tions operators and manufacturers of cess is distributed over several ma- area network, it is done having a specific computers, software and telecommunica- chines; purpose in mind. Unspecified or wrongly formulated or focused purposes are com- • The engineering viewpoint specifying mon sources for bad system designs and 1 ISO = International Standardisation how the distribution takes place and wasted investments. Hence, a good enter- Organization; IEC = International which mechanisms are required in prise specification is the same as having Electrotechnical Committee; ITU = order to do this; a good and well formulated strategy for International Telecommunication • The technology viewpoint specifying the system. A properly defined purpose is Union. how the system is actually built in the best guiding principle for the modell- 2 CORBA = Common Object Request terms of machine configurations, oper- ing, the design and the implementation Broker Architecture. ating systems, programming languages that follows. It is, for example, common and systems for operation and manage- in telecommunications that customers 3 TINA = Telecommunications Informa- ment of the final system. ask for a specific technology, say, an tion Networking Architecture.
24 Telektronikk 3/4.1998 expensive fixed network ISDN imple- It is in my opinion an enigma how the Every enterprise has a set of policies of mentation, rather than a solution, for year 2000 problem was overlooked or its own and has, generally, to meet poli- example, anything that can provide, as ignored by both those who needed the cies and regulations set by external bod- soon as possible and at the lowest pos- systems, those who specified them, and ies. Examples of such policies can be sible price, telephony and data trans- those who made them. This is the best related to information security, openness, mission at a rate at or above 9.6 kbit/s. example we have of the consequences of protection of personal integrity, protec- This requirement may be solved in sev- a complete unconsciousness concerning tion of third parties, rules for co-opera- eral ways, for example, providing ISDN, the enterprise view of the systems to be tion with other enterprises and business traditional telephony with modem, purchased. And the problem is universal. conditions. The enterprise viewpoint is mobile telephony or ATM depending on This illustrates how little concern enter- the right place to identify which policies which technology is available. Defined prises are putting in a proper strategy for and regulations, internal and external, in this way, all these solutions are accept- the systems they make their entire enter- will have impact on the design. At all able for the user. The solution may even prise depend upon. stages of modelling, design and imple- be enhanced or replaced as technology is mentation, it should be verified that poli- developing, for example, replacing an Every enterprise has an environment and cies are not violated. The sector specific early telephone solution with ATM when every system within that enterprise has to regulation of telecommunications is a the technology becomes available. interact, directly or indirectly, with that good example of the impact of the en- environment. These interactions need to vironment. It also shows how difficult it Take GSM as another example. The de- be identified and modelled into the sys- is to understand this interaction for com- fined purpose of that system was to pro- tem design and implementation. The plex systems like telecommunications vide Europe with a land mobile system more indirect these interaction are, the (see [3] for more details). which allowed users to roam all over more difficult it is to identify their impact Europe, maintaining all subscribed ser- on the design. Similarly, every new sys- The final point in the list above concerns vices independently of point-of-access, tem introduced in an enterprise will technical issues like performance, relia- and having only one point-of-subscrip- impact future evolution of the techno- bility, dependability and so on. Any such tion and one point-of-billing. The defini- logical infrastructure of the enterprise. requirement may have a dramatic impact tion of the purpose thus contains infor- Therefore, modelling should allow for on implementation choices and cost since mation on the geographic area where the open ended design, i.e. design that can they may require duplicated groups of system should be available (Europe), and easily be extended or fitted into a new equipment and expensive, high quality conditions that have to be met in the geo- environment. The modelling principles parts and components in order to meet graphic area irrespective of network of the TINA Consortium are based on reliability specifications, and much spare operator, service provider, customs barri- this type of approach because telecom- capacity of computers in order to meet ers or legislative constraints in each munications is one of the fastest chang- performance requirements for stochasti- country involved. Among the first items ing technologies today. The openness cally varying loads. This is particularly to be solved in order to meet the purpose of telecommunication systems is often true for telecommunications equipment was to sort out issues like free passage of manifested by a web-like interconnect where the impact on the society of fail- mobile stations across national borders, structure of applications among a large ures may be tremendous [3]. However, in particular equipment containing en- number of individual and very dissimilar they have usually little impact on modell- cryption devices. If these problems were enterprises. Closely attached to this web ing and design. not solved, there would be no reason for are also the telecommunications and designing the system. This is an example computer infrastructures of the users. Much of the success of system develop- of a strategy which goes far beyond the ment lies in formulating clear and rele- system design itself. Again we may use GSM as an example vant enterprise requirements, and stick in order to illustrate the dependencies to them all the way down to implementa- Already in the planning phase, i.e. at between systems and how they may be tion. enterprise viewpoint, it is often important handled. The embedding requirements to state how the system should be admin- were 1) that the GSM system should fit The remainder of this essay will contain istratively embedded in the enterprise, seamlessly into the existing fixed net- some general enterprise models for i.e. defining responsibilities, who has work, and 2) that no redesign of the latter telecommunications. These models are access to the system, who can do what would be necessary in order for them to also inherent in every system using to the system, etc. This may for instance interwork properly. Both requirements telecommunications. From an enterprise define the security levels the system has were achieved. However, some of the point of view the enterprise model for a to support. Normally, a new system is features of GSM were then designed in a distributed application using embedded in an existing environment way which was not optimal neither for telecommunications for connectivity may within the enterprise, requiring guide- that system alone nor for the fixed net- be visualised as in Figure 1. lines concerning how it should interact work. The latter has later adopted some with this environment. If this point is not of the mobility features of GSM in order Every distributed system needs the sup- addressed early and at a high level, it is to support services like universal mobil- port of telecommunications. This estab- likely that the installation of the new sys- ity. In this case there was an enterprise lishes some primary concerns. The enter- tem will require expensive adaptation and requirement but it turned out to be sub- prise model of the application must inter- cause delays. The lack of a clear view on optimal in the long run. This just illu- act with the environment of the enter- embedding problems seems to be a com- strates how difficult this type of strategic prise itself and with the telecommunica- mon cause that so many systems only specification is, and how easy it is to end tions system supporting the distribution partially fulfil their intended purpose. up with bad solutions. of the application and its interactions
Telektronikk 3/4.1998 25 the field act as if that is the case) but distributed application. This corre- where the main driving forces are rather sponds to the specifications of the Enterprise model to do more business in the primary area information, computation and engin- for application (trade), to be more competitive and to eering viewpoints. This specification influences influences reduce costs and increase profits. is beyond the scope of this essay. The influences interested reader is referred to [4]. Enterprise model Telecommunications will be a support for telecommunication • The service architecture uses the system for many different applications. It methodology of the computing archi- will partly be general, that is, applicable tecture for defining and implementing Environment for many different applications, or telecommunication services. This specifically designed for one application. architecture contains the concepts of Figure 1 Relationship between This is what makes telecommunications telecommunications. Here we have to enterprise models so difficult and requires that separate bear in mind that information technol- enterprise models are developed for ogy and telecommunications are con- telecommunications which may be inte- verging technologies which means that grated with enterprise models of very dif- the distinction between them may dis- with the environment. Some of the inter- ferent applications. The skill to do this appear in the future. This applies in actions with the environment may be may become a core competency in particular to the software used in such direct or via the telecommunications sys- telecommunications companies in the systems. However, the service archi- tem. When specifying a model it is then future. Few are clever at doing this today. tecture also describes such items as essential to define three sets of interac- quality of service (bit or word error tions and their possible interdependence. This is also a challenge for the telecom- rate, response time, transfer rate, infor- The latter is probably the most difficult munications system designer. The system mation loss probability, synchronisa- task of the design and probably often must contain attributes which are general tion), fault recovery, pricing principles overlooked. In particular, it may be diffi- in the sense that they may be used by and service interaction handling. These cult to see the secondary relationship many enterprises and, at the same time, are elements which are absent from imposed by the telecommunications sys- be adaptable to any specific enterprise most specifications of information sys- tem on the environment and the impact and its environment. The enterprise mod- tems but are of utmost importance in of any limitations in the telecommunica- els which are presented below for properly designed telecommunications tions system itself and how such limita- telecommunications are general and de- systems. tions may be bypassed using other rived with these aspects in mind. means. Designers of applications are • The network architecture defines the often taking telecommunications for computational issues of the infrastruc- granted and do not incorporate it as an 3 Telecommunications ture, that is the access network, the integral part of the overall design. For Architecture transport network and the platforms example, when devising a system for (exchanges, servers, databases) respon- electronic commerce, different forms of The TINA Consortium has defined an sible for routing, processing and ser- telecommunications are needed in order architecture for telecommunications vice handling. One important point to meet the overall goal: web-like sys- which separates the system into five here is that the architecture distin- tems for browsing of catalogues, trans- interrelated architectural entities as guishes clearly between services and action systems for payment, security sys- shown in Figure 2. networks. This has two important con- tems for avoiding fraud and supporting sequences. First, services and applica- • The overall architecture contains all non-repudiation, transport channels for tions can be defined independently principles applicable to all designs. delivering electronic goods, etc. This is of which network they are run on. Sec- This contains all methodology and not a business area where telecommuni- ond, the fast evolving services and the principles of the ODP viewpoints, cations is at focus (though many actors in application technology become in- including the enterprise viewpoint. dependent of the more static network As explained above, the same overall technology. The difference in evolu- architecture can be used for both tionary speed of the two will probably telecommunications and the applica- become more pronounced because of tion requiring it or, to put it in other Overall the low entrance threshold for services words, applications should be specified architecture and the high investment requirements using similar tools and methods as for for networks [3]. telecommunications in order to Computing increase the benefits of the specifica- • The management architecture consists architecture tion and to identify the influences of applications required for managing, between the enterprises involved. monitoring and maintaining the other elements of the system. The architec- • The computing architecture is a subset ture is based on the telecommunica- of the overall architecture containing tions management network (TMN) Service Network Management all principles for modelling, designing defined by ITU and ISO. This means architecture architecture architecture and building the distributed system. that TINA specifications use different This architecture is independent of methods for defining services and whether the application is a telecom- management. Figure 2 Hierarchy of architectures munications application or a general
26 Telektronikk 3/4.1998 System integrator As indicated in the figure, several rela- tionships exist between the architectures. 3rd pty It is important to note that all architec- Consumer Retailer service tures must be in compliance with the provider computational architecture. This means simply that the modelling principles should be used consistently when defin- Connectivity provider ing services, transport and management. Managing When defining a service or an applica- relationships tion, an abstraction of the network will Clearing house normally be needed in order to show how the service or application interacts with For a single consumer there may be the network resources. This abstraction is many instances of all other entities given by the network architecture. Note that this abstraction may also hide which Figure 3 The information service supermarket elements are realised as hardware or soft- ware or which technology is used. This is what enables the separation between ser- vices/applications and networks. tivity and traffic capacity obtained from tual reality delivered jointly by a number 4 Business Architecture connectivity providers and combining of operators. The reasons are, of course, them with own or third party services 1) that retailers and system integrators An important part of the enterprise view- such as customer care, provision of con- may put together any set of business point is to understand the business rela- tent, storage of content, design of content interests, and 2) that a digital network tionships that are required. This is partic- (web pages), and support of transactions. may pass any type of information pro- ularly important for telecommunications The structure may become so complex vided there is enough bandwidth. The since the international network, owned that system integrators are required to latter may mean that all services and by a large number of companies, offers keep the structure together. Note that this applications may not be available to full connectivity among all its customers structure will be able to offer any service everybody everywhere. Note that there and thus requires co-operation between from plain telephony to the most ad- may be more than one entity of the same all these companies. In addition, telecom- vanced multimedia services based on vir- type (retailer, system integrator, third munications is now changing character in that new markets, services and network technologies are evolving and that sev- eral new types of companies offering different aspects of telecommunications are entering the market [3]. Platform operator
A business architecture for telecommu- Connectivity provider nications can be built in many ways. In Telenor we have developed a business Access provider Transport provider model consisting of ten atomic roles plus the content provider [3]. The roles are atomic in the sense that it is meaningless to decompose them further and that any Infrastructure owner company in telecommunications can be present in a single role or be composed of any set of such roles. The TINA Con- Consumer interface sortium has proposed a similar but simpler model, the information service supermarket model [5]. We use this Retailer model here because it is well suited for our purpose which is to identify a few System integrator primary domains into which the system Subscription/ may be subdivided. The model of customer care/ Telenor is better suited when designing billing manager Hosting business models and competition scenar- ios. The model is shown in Figure 3. Sales The model consists of the following main elements. The customer purchases ser- vices from a retailer reselling connec- Figure 4 Decomposition of roles
Telektronikk 3/4.1998 27 party service provider, connectivity providers and clearinghouse) offering services to one customer. System In a structure like this it may also be Business relationship integrator possible to combine telecommunications 1 services with other businesses, for ex- 1 1 ample combine electronic commerce 8 1 with physical delivery of goods and 7 advanced transport logistics where, for 2 7 3rd pty example, Rema 1000 sells the goods Consumer Retailer service electronically and Pizza Hut delivers provider them at the door of the purchaser, and there may be a third company computing 4, 6 the delivery routes (travelling salesman 3 3, 5 3 algorithm or the like) and thus reducing the cost of delivery. 9 Connectivity provider The purpose of the clearinghouse may be to establish trust among the parties, act as a notary public for proving that transac- 10 10 tions have taken place (non-repudiation), offer anonymity and erasure of electronic traces, and perform security functions Clearing such as integrity, certification and house Figure 5 Business relationships encryption. The clearinghouse may also offer support for complex transactions.
Figure 4 shows another way of looking at the business system. Here the retailer and the connectivity provider have been di- vided into several roles. The retailer roles The model is becoming even more com- vironment. This figure also indicates that may be sales, hosting (for example offer- plex realising that the consumer may be there may be more than one role of each ing Internet servers and web hotel), man- different things in different contexts: a type in the system and that interaction agement of customer relations (subscrip- single person or a company, a machine, a with different parties may be indirect. tion, billing, customer care) and system software entity, or a combination of these. The customer may, for example, receive integration. The latter represents the The telecommunications service may be a service produced jointly by one third complex role of integrating services, net- a primary service, that is, it is the final party service provider and several con- works and content from a number of product. Person-to-person telephony is an nectivity providers but offered to the cus- independent operators. The connectivity example. The service may also be tomer through an interface of type 1 by a provider may be a collection of access secondary, that is, delivered as part of system integrator. The customer will in providers offering access on copper, fibre another product. In electronic commerce this case only see the system integrator or radio; transport providers owning telecommunications is only one (but nec- and not the parties actually producing the cables, fibres, satellites or multiplexing essary) component of the offered product service. equipment; platform operators running and as such a secondary component in the machine infrastructure consisting of the offer. Telecommunications may also The relationship between two business servers, exchanges and computer nodes; be a ternary product in the sense that it is roles may exist over a long period of and infrastructure owners owning build- used to produce a product but is not a time (for example, so long as a subscrip- ings, ducts, masts and other basic infras- visible part of the product. Telecommu- tion exists) or only for a single call or tructure required by the others for build- nications used to enable payment trans- part of a call. ing the system. This also shows how we actions in electronic commerce may be may easily superimpose the model in [3] regarded as a ternary product since in The business relationship may be defined on top of the TINA model. The customer principle, it can be exchanged with any in terms of attributes. Take the relation- may have business interfaces to these payment scheme without altering the ship between the customer and the retailer extended structures at none, one or sev- product as seen by the consumer. (marked 2 in Figure 5) as example. The eral points. However, physical interfaces attributes may be: may exist independently of business As is seen in Figure 5, the model requires • Set of services and applications; interfaces: a customer may have only several types of business relationships the one business interface with a sales between the various parties. The figure • Billing conditions and prices; organisation. The physical access inter- indicates that as many as ten different • Conditions related to dialogue initia- face with an access provider is an in- business relationships may exist, each of tion; direct part of this business relationship. which characterised by a set of attributes and rules. Telecommunications compa- • Set of business domains required; nies therefore see a very complex en-
28 Telektronikk 3/4.1998 • Binding requirements; Split interceptor • Quality of service management.
In an object oriented model these may Administrative Administrative be attributes of the relationship or the domain 1 domain 2 objects (or roles) it interconnects. This item will be considered in [6].
From a strategic point of view it is important that all these relationships are specified for a given configuration. It is Technology In-line interceptor Technology particularly important to verify that rela- domain 1 domain 2 tionships and attributes are not forgotten in this process. Figure 7 Interceptors 5 Domains and Domain Interactions • Agents acting on behalf of another • Split interceptors interconnect admin- One of the most important aspects of domain or entity of that domain; istrative domains. The interceptor is distributed systems is the concept of responsible for ensuring that security • Gateways performing translations or domain. Every public or private telecom- and other policies are met when two conversions between domains; munications system is open, meaning domains are connected together. As that it interacts with other systems in • Monitors surveying domains or the name suggests, the interceptor order to connect customers and provide elements of the domains. does not take part in the information joint services to them. In this respect the exchange. It is only needed when system is divided into domains indicating Another way to look at interceptors is in security and policy information are ownership. Such domains are called terms of which type of domain they inter- exchanged between the domains. It administrative domains. A business role connect. In this case two types of inter- may, however, perform continuous as described in Chapter 4 may be an ceptors may be defined (see Figure 7): monitoring functions, for example in administrative domain, or an administra- tive domain may consist of several busi- ness roles if they are owned by the same company. In other cases two systems built to different technologies are re- quired to interwork; that is, to provide joint services to their customers. In such Administrative cases the systems belong to different domain technological domains. These are in- (In-line interceptor) dependent of ownership. Technology Technology The domains need to interact in order to domain IW domain perform their joint task. The interaction ISDN ATM need not be symmetric. The configura- tion of two interacting domains is shown in Figure 6. The possible asymmetry is IW = Interworking indicated by assigning different sets of rules (12 and 21) for the two directions.
The interaction between domains takes GSM place via interceptors. The interceptors Technology are abstractions of the interfaces between domain the domains and may act as: Administrative Administrative domain domain Operator 1 Operator 2 Split Rule set 12 interceptor Domain 1 Rule set 21 Domain 2
Figure 6 Domain interactions Figure 8 Examples
Telektronikk 3/4.1998 29 Split interceptor The domain concept may be used in the business model. An administrative Administrative Technology domain may span several business roles, domain 1 domain 3 or several business roles (of the same Technology type or of different types) may belong to domain 3 Technology Administrative the same technology domain. This is domain 1 Technology domain 3 illustrated in Figure 10 showing two Administrative administrative domains each including domain 2 domain 2 two technology domains. One of the In-line technology domains is common for the interceptor Split two administrations. Each administrative interceptor domain consists of several business roles. The figure also shows that there may be business roles not containing a technol- In-line ogy domain. Most retailer roles may be Figure 9 More complex interceptor of this type. Similarly, one business role domain structure may offer several technologies, for ex- ample a connectivity provider having both ISDN and ATM networks. The example does not show the interceptors. However, there will be a split interceptor between the two administrative domains and in-line interceptors between technol- order to detect policy violations. The i.e. two administrative domains are inter- ogy domains of different types (if the split interceptor may be of type agent connected. A split interceptor is needed technologies are interconnected). or monitor. where the interceptor essentially takes care of the security between the system • In-line interceptors interconnect tech- A final point on domains concerns the and the mobile station, location updating nology domains for interworking and external visibility of the roles of the and routing of calls to the mobile station. conversion. The in-line interceptor business model. A consumer will, for may be of type gateway. instance, only see the role with which Domains may consist of domains as there exists a business relationship and a shown in Figure 9. In the example Figure 8 shows two examples. In the first ‘federated’ role which may be called the administrative domain 1 consists of two example two technology domains of the telecommunications system with which technology domains interacting via an in- same administrative domain are intercon- there may exist a technical relationship line interceptor. Technology domain 3 nected. One domain is built with ISDN (access, connectivity). The federated role consists of two administrative domains technology and the other with ATM tech- and the relationship with it will then have interacting with technology domain 2 via nology. The interceptor will in this case only the attributes which are important an in-line interceptor. Between the admi- be of type in-line interceptor and perform for the technical interconnection of the nistrative domains there are split inter- interworking of the two technologies. In consumer and the telecommunications ceptors. It is easy to envisage much the other example there is only one tech- system. It is unimportant whether the more complex situations than this. nology domain (GSM) but two operators; federated role consists of a single entity or is composed of several connectivity providers and third party service pro- viders. In general, it is possible to feder- ate roles in this way in order to hide com- plexity. For the customer it is un- important how a service is composed or who provides individual parts of it. Business role 1 Moreover, various sets of different roles may produce the same service. The result is the simple domain structure shown in Business role 2 Business role 3 Business role 4 Figure 11 where the consumer and re- tailer roles are kept in order to show the subscription relationship but all the other Technology Technology Technology domain 2 roles have been federated into the domain 1 domain 3 telecommunications domain.
Another model that will come out handy Administrative Administrative in other contexts, for example when dis- domain 1 domain 2 cussing middleware and transparency for mobility [7], is shown in Figure 12. Here Figure 10 Combination the retailer domain and the telecommuni- of business model and cations domain have been federated and domain model the new terminal domain has been intro-
30 Telektronikk 3/4.1998 Retailer domain The activity between a user, or rather the supporting terminal domain, and the telecommunications domain is called a Consumer domain Telecommunications domain user session. Activities between two sub- entities of the telecommunications domain or between telecommunications Figure 11 Federated domains domains are similarly called provider sessions. The overall activity is simply called a session. Figure 13 shows a session consisting of user sessions and provider sessions. Using this notion, communications between three parties can be illustrated as in Figure 14. The three interworking telecommunications User domain domains are connected by one provider Usage relationship Subscription relationship session in the model.
Connection relationship There are several important points related Terminal domain Telecommunications domain to sessions. A session must be estab- lished before any communications can take place. The establishment of the ser- vice may be regarded as a separate sub- Figure 12 Other federation session called an access session. Sessions are independent of one another. The set of sessions of a telecommunication ser- vice or application may change with time; that is, the number of participants and the type and number of resources duced. The consumer domain has been • The terminal is connected to the net- allocated to the process may vary. Again redefined as user domain. The terminal work. In some realisations it may be an organising type of session called com- domain will then contain all user func- regarded as part of the telecommunica- munication session may be introduced in tions and support the technical interface tions domain; in other applications it order to provide an abstract view of the with all other roles. The relationships may be part of the consumer domain. connection or interdependence between between the three domains consist of In some applications such as mobility different sessions; that is, providing a a subscription relationship between the systems, it is advantageous to regard it connection graph for applications like user domain and the telecommunications as a separate domain [7]. the one shown in Figure 14. system domain; a connection relationship between the terminal domain and the In the federated model there may be Instead of talking about end-to-end appli- telecommunications system domain; and split interceptors between two and two cations it is better to refer to incoming a usage relationship between the user domains since they are all administra- applications and outgoing applications domain and the terminal domain. The tively separate. Between the terminal where an outgoing application is initiated model of Figure 12 is thus not strictly a domain and the telecommunications by the user, while an incoming applica- role model but rather a technical one domain there may also be an in-line tion is initiated by an entity in the tele- bringing forward three important ele- interceptor, for example a modem be- communications domain or by another ments: tween a PC and the telephone network. user, that is, by an outgoing application. A two party call will then consist of one • The terminal and the network may outgoing application and one incoming belong to different administrative 6 Sessions and application interconnected with each and/or technological domains. In a other. The incoming application is in this general model we may assume that Applications case started by the outgoing application. the three domains of the figure are In this chapter the domain model of Fig- An example is shown in Figure 15. In a different both from an administrative ure 12 will be used. The telecommunica- and technological point of view. tions system offers services which • A user is normally not a technical together with other processing make up device but needs a terminal on which an application. This is the same as saying to run the local applications of the ser- that telecommunications should not be, vice. Note that a user may not only be as it often has been, regarded as a stand Session a person but also a hardware device or alone product but as a component of an User Provider a software entity. However, business application also involving other resour- sessions sessions information such as user service pro- ces. Applications like electronic com- file and service restrictions may be merce, distance learning and electronic required (supported by the subscription banking are examples of applications Session as a composite relationship between the user domain where most of the application is related and the telecommunications domain). to things other than telecommunications. Figure 13 Session and sub-session
Telektronikk 3/4.1998 31 User User domain domain be unsolved even though the computer Telecom may otherwise work according to spec- Telecom User Terminal domain ifications); domain session domain User Terminal Provider session • Small changes at the enterprise level domain sessions may require major redesign of the User system. Telecom domain domain User Therefore, it is essential that much effort Figure 14 Three party session is put into this part of the specification Terminal interaction as example and not, as has apparently happened in domain several cases, be ignored completely. ODP offers a top down approach to design where each viewpoint depends on the other in an hierarchical order. As we have seen above, the topmost element of this hierarchy, the enterprise viewpoint, User User may also be broken up into several areas domain domain which may be specified independently. Telecommunications domain From a strategic point of view this is unquestionably the most important part Terminal Outgoing application Incoming application Terminal of the specification since it will set the domain domain requirements for everything else which is said about the system. Figure 15 Applications References
1 ISO/IEC/ITU. Open distributed pro- cessing : reference model : founda- particular communications event there of complex systems. By using the pro- tions. ISO/ICE Rec. 10746-2, ITU-T may be any number of incoming and out- posed models for the structure of the Rec. X.902, 1996. going applications. overall architecture, the business rela- tionships, the domain structure and the 2 ISO/ICE/ITU. Open distributed pro- There are several examples where one concepts of session and applications it is cessing : reference model : architec- of the users is the network (or, in order possible to reduce the complexity of the ture. ISO/IEC Rec. 10746-3, ITU-T to be formal, the telecommunications system or divide it into manageable Rec. X.903, 1996. domain): the wake up service is an in- pieces. Even though the methods seem coming only application; the location simple and almost trivial, this is in fact 3 Audestad, J A. Telecommunications updating service in the GSM system is the most complex part of system design. and complexity. Telektronikk, 94, an outgoing only application. One very important reason for this is that (3/4), 1998, 2–20. (This issue.) issues related to enterprise specifications are close to the decision making part of 4 Audestad, J A. Distributed process- 7 Enterprise Specifi- the enterprise. This implies that ing in telecommunications. To be published. cations and Strategy • The impact of errors may be great; The outline given above of the enterprise • If the enterprise specification is wrong, 5 Mulder, H (ed.). TINA business viewpoint for telecommunications offers it does not matter whether the design model and reference points. TINA-C a reasonable approach to the modelling is correct (the year 2000 problem may Baseline Documents, 1997.
6 Audestad, J A, Kjæreng, A, Mahieu, L. An object oriented model of telecommunications business. Telek- tronikk, 94, (3/4), 1998, 54–61. (This Jan A. Audestad (56) is Senior Advisor for the Corporate Management of Telenor. He is also Adjunct Professor of issue.) telematics at the Norwegian University of Science and Technology. 7 Do van Thanh. Mobility as an open distributed processing transparency. e-mail: [email protected] PhD thesis, University of Oslo, 1997.
32 Telektronikk 3/4.1998 Agent Nets: a methodology for evaluation of enterprise strategies in information economy
ALEXEI A. GAIVORONSKI
In this paper we address new challenges Institute, ETSI, and in particular EPIISG, • Transition from a highly homogeneous which are now facing corporate strate- the follow up of Strategical Review user population to a highly diversified gic planning in the telecommunication Committee number 6 of ETSI. The population of users of information industry. These challenges are the con- objective of these bodies was to under- services, according to their social posi- sequence of two interrelated phenom- stand the problems involved in defining, tion, habits, mobility, etc.; ena: the emergence of the information standardising, engineering and imple- • Transition from a tightly controlled industry as a result of convergence menting the so-called Information Infras- monopolistic environment with fairly between the telecommunications, com- tructure with specific focus on the role of well defined and few industry roles puter and content provision industries; the telecommunications industry in this basically reduced to a network pro- and the accompanying deregulation infrastructure. In particular, SRC6 devel- vider and equipment manufacturer to process. Traditional approaches to the oped the descriptive Enterprise Model of enrich the ever expanding variety of evaluation of enterprise strategies fail the Information Industry with the aim of industry roles with blurred boundaries to produce adequate results because evaluating regulatory and industrial sce- between the provision of structural and they lack capabilities to represent rapid narios [8]. It became evident that in order infrastructural services and a multitude changes, non-stationarities, uncertainty to enhance normative activities it is of loosely regulated and competing and complex relations between a multi- important to support them with adequate players, each assuming different roles; tude of players, which characterise the modelling tools, which would permit new industry. evaluation of enterprise and regulatory • Transition from centralised decision strategies in the new industrial environ- making to distributed decision making. Our answer is to utilise a new method- ment. The aim of the project described in This occurs on two related levels: on ology for evaluating enterprise stra- this paper was to select, adapt and further the level of industrial relations tegies, which is based on the emerging develop such tools. In the early stages of between different economic agents theory of distributed multi-agent sys- this project we benefited greatly from the and industry players, and on the level tems. It takes the best features from participation and encouragement of Dr. of telecommunication networks which more traditional approaches like sys- Mario Bonatti from ITALTEL who at are being transformed from a relatively tems dynamics, and enhances them that time served as Rapporteur of EPIISG simple uniform structure to a federa- with the latest advances from modern Working Group on the Specification of tion of interacting networks equipped economic theory, operations research, Enterprise Interfaces. with different levels of intelligence; game theory and discrete event simula- tion. In order to emphasise the net- • Emergence of a multitude of small and We started from the premise that one of work structure of modern industries it medium sized economic agents in an the most important phenomena of con- is called Agent Nets. It is specifically industry which is accustomed to big temporary economy and society is the designed to represent complex systems players; unfolding of the information industry as made up by independent entities called a result of convergence of the telecom- • Network nature of industry competi- agents which transform and exchange munications, computer and information tion and collaboration patterns; information and other resources mak- content provision industries. This process ing independent and co-ordinated • An increased importance of dynamics brings about radically different relations decisions on the basis of incomplete and uncertainty – the state of equilib- and organisational forms within the information about the state of the rium, if ever reached, is determined by industry as well as radically different whole system and the actions of other the history of the process, and some- relations between producers of informa- agents. In this paper we discuss the times by decisions made at a relatively tion, providers of telecommunication applications of Agent Nets for the eval- early stage. infrastructural services, and users of uation of enterprise strategies, present information services. This process is a mathematical description of Agent Traditional economic and organisation accompanied by rapid organisational and Nets, describe an Agent Net simulator theory has difficulties confronting these institutional changes, which pose new – MODAGENT – created for simula- problems because it mainly addresses the challenges for both industry regulatory tion of multi-agent systems, and pre- systems in equilibrium and has much less bodies and for individual telecom enter- sent a case study dealing with the mod- to say about the systems in the state of prises, which need approaches and tools elling of industrial relations in the change and in conditions of uncertainty. for evaluating different regulation and information industry. We have found that a majority of tradi- enterprise strategies under various tional modelling and decision support development scenarios. tools suffer from the same drawbacks. 1 Introduction In order to be useful such planning tools For this reason we decided in this project should possess capabilities of reflecting In this paper we describe current results to focus on the emerging theory of dis- the following phenomena: of our ongoing research project whose tributed multi-agent systems, further de- objective is to contribute to the under- • Transition from technology driven veloping this theory and its application to standing of information economy and industry to market driven industry (at the telecommunications industry. This develop methodologies and tools which least to a much larger extent than pre- methodology utilises the best features can be used by players in the industry to viously); from more traditional approaches like evaluate their strategies in a time of rapid systems dynamics, and enhances them • New economic phenomena which technological and organisational change. with the recent advances in economic guide the creation and distribution of This project originated in connection theory, operations research, game theory, information products; with standardisation activities of the and discrete event simulation. European Telecommunications Standards
Telektronikk 3/4.1998 33 Soon after starting our project we real- In the context of the information industry • Network structure. All these systems ised that modelling distributed systems the modelled system is the information can be represented as graphs with of this complexity is very much an open economy, and agents are the enterprises agents being the vertices of the graph. question, although several promising and business units involved in creating, The edges of the graph represent approaches are starting to make the first producing and distributing information exchanges and interactions between steps in modern economics, computer products, network providers, consumers agents. For example, in the case of science and mathematics. Therefore, we of information products, government and telecommunication networks these had to organise our research in three regulatory agencies. Such agents make edges correspond to the links between interrelated directions: decisions about the consumption, trans- nodes, while in the case of information formation and exchange of information, economy the edges are supplier-con- 1. Development of a mathematical products, services and other resources, sumer relationships. methodology for modelling multi-agent expand their production facilities, and • Transformation and exchange of systems. formulate their strategies in order to resources and flows. In all cases agents 2. Development of an object oriented achieve specific aims. These decisions possess internal resources, transform modelling system based on this method- are taken in an asynchronous and dis- input resources and flows, in- and out- ology. tributed manner. Agents may combine put resources and flows, and exchange 3. Applications to the modelling of different roles within economy, like con- traffic flows and resources. In the case telecommunications systems and infor- tent provision, brokerage, and delivery of of the network nodes such internal mation economy. information service. resources are represented by process- The rest of the paper is organised as ing capacities and buffers, the input follows: In Section 2 we discuss proper- A more detailed analysis of the informa- flows are received packets and in- ties of multi-agent systems and related tion industry as a multi-agent system can bound connections, the output flows modelling challenges. In Section 3 we be found in [6, 7, 8]. are outgoing packets and connections. formally describe the notion of Agent In the case of enterprises internal re- Net and discuss some of its properties. We use the word agent in describing sources are production capacities Section 4 is dedicated to an informal dis- these entities in order to emphasise their which transform input resources into cussion of Agent Nets. The simulation capability to make independent deci- products and services. In the case of system MODAGENT developed for sim- sions, communicate with each other, and end users internal resources are termi- ulating Agent Nets with application to exchange different commodities. Histori- nals, while input resources are connec- modelling of information economy is cally this word is used extensively in tions, products and services. Agents presented in Section 5. This system is economic modelling to describe eco- exchange flows and resources. Ex- applicable also for modelling of other nomic subjects who make independent amples of such exchanges include multi-agent systems found in telecommu- decisions. More recently the notion of exchange of products for money, nications. Section 6 is dedicated to the ‘intelligent agents’ was introduced to exchange of signalling between net- case study from modelling of relations describe computer programs operating work nodes while establishing a con- between network providers and providers in a distributed and networked reality nection, and exchange of information of information services over telecommu- like the Internet and acting on behalf of a for information. In what follows we nication networks. human making intelligent decisions. In consider as resources all ‘passive’ our modelling of systems composed of commodities transformed and ex- agents we draw upon both of these changed by agents, including resources 2 Distributed multi-agent notions. in common sense, but also all kinds of systems products, services, information and In all of these cases a number of success- traffic flows. The variety of distributed systems can be ful approaches for modelling individual described as a collection of relatively agents exists. Modelling systems of • Distributed asynchronous decision- independent units, here called agents, agents, however, is a much more difficult making and control. All agents make which process information, traffic flows task and existing approaches give only decisions which involve assignment and other resources; exchange informa- partial solutions for relatively simple of input resources, treatment of input tion, resources and money; make decen- problems. Still it is a very important and output resources, selection of part- tralised decisions which affect the local problem for the design and engineering ners for exchange. These decisions are state of the networks. Taking telecommu- of distributed systems. In this paper we made in an independent and asyn- nications as a reference, we can see that consider an approach for modelling chronous way, although they can be co- in the case of the global area network multi-agent systems which we call Agent ordinated. Examples of such decisions such agents are the network nodes which Nets. are routing, admission control and con- exchange traffic flows and signalling and gestion control algorithms of telecom- make routing and congestion control We start by observing that although munication network nodes; production, decisions. In the case of local ATM net- multi-agent systems mentioned above are development and purchasing policies works such agents are local exchanges very different by nature, they possess the of enterprises; and selection of service with admission control units negotiating following common features, which providers by end users. In their selec- resources with users and supervising the suggest a development of common tion of decision strategies the agents user behaviour, while the users them- methodological approach for their mod- seek to follow their individual criteria selves can be considered another type elling: and aims which are generally not co- of agents. ordinated with each other. Therefore, the design of multi-agent systems can-
34 Telektronikk 3/4.1998 not be reduced to simple global optimi- posed from fairly homogeneous agents where computers or ‘agents’ of partici- sation criteria, like minimisation of with similar behaviour patterns. pants can simulate trading processes total costs or maximisation of properly Instead, we needed the capabilities to and reach equilibrium before making defined ‘public good’. model a rich variety of relations where real decisions. The existence of such a the same agents can compete in one market system with agreed rules on • Dynamics and non-stationarity. Multi- field and collaborate in another over- how it should function and an anony- agent systems can exhibit widely dif- lapping field, assume different combi- mous decentralised exchange of neces- ferent dynamic behaviour due to rich- nations of industry roles, and possess sary information may allow us to reach ness of positive and negative feedback different knowledge of the state of the efficiency for incomplete markets. The usually present in such systems. They whole system. This complexity leads design of multi-agent market systems can have many equilibria and switch in to a multitude of positive and negative requires special decomposition tech- catastrophic manner between them. feedback in the system, which under niques and the use of (often stochastic) They can exhibit totally chaotic be- different values of system parameters optimisation techniques. haviour even in the absence of random can lead to different equilibria, and disturbances. Design based on the even chaotic behaviour. Even without We confront these modelling challenges study of steady state behaviour is not chaos the presence of multiple equi- by utilising and further developing the adequate because in many cases such libria leads to catastrophic behaviour; emerging theory of distributed multi- systems change constantly. Examples i.e. in certain points the system abrupt- agent systems. In particular, we consider of such non-stationary changes include ly switches between different equilib- a new modelling and simulation method- the rapid development of mobile net- ria with an arbitrarily small change of ology called Agent Nets. It is a network works and the explosion of the Inter- system parameters. An important like structure with agents being the ver- net. The large part of traditional eco- objective here would be to define tices of oriented graphs whose edges nomic theory and modelling is centred ‘robust’ patterns and regions of sta- define the exchanges between agents. on perfect markets in the state of equi- bility. This structure is associated with appro- librium. In such systems the operation priate resource space with agents trans- of market forces smooth out disturb- • Bounded rationality. Traditional eco- forming and exchanging resources from ances introduced by uncertainty and nomic theory assumes that economic this space. Each vertex of the graph is leads the system to an ergodic state of agents are perfectly rational and their equipped with the set of transformation equilibrium. In the case of rapid tech- behaviour is governed by maximisa- functions and strategies. The state of the nological change this is no longer the tion of certain utility functions. Be- vertex (agent) is defined by the vector of case because relatively small disturb- sides these ideals, the ‘best case’ internal, input and output resources. ances and decisions with little imme- scenarios, we included in our system There are dynamic relations which guide diate impact can have self-magnifying some other models of agent behaviour the evolution of the state of the agents properties due to the positive feedback which assume the bounded rationality similar to those considered in the theory present in the system (technologies of agents, i.e. that their decision ac- of Discrete Event Dynamic Systems with increasing returns) [4]. This leads tions result from the set of heuristics DEDS) [9, 12, 17, 21, 28, 30]. Recently to non-ergodicity of the system, which which vary according to the changing developed chapters of this theory which requires the modeller to shift the em- of information patterns, environment deal with simulation models of asyn- phasis from stationary to transient and goals [3]. Such heuristics may chronous systems and with interplay behaviour. cause instability and they are con- between simulation and optimisation are stantly being evaluated against ob- • Uncertainty. The lack of ergodicity particularly relevant in this context. tained results, and new heuristics are increases the importance of adequate generated. treatment of uncertainty present in the We draw upon the experience accumu- system. There are two levels of uncer- • Imperfect markets. Market mecha- lated recently in modelling and applica- tainty present in the system: external nisms are traditionally analysed under tion of network like structures, in parti- uncertainty represented by demand the strong assumption that all actions cular Petri Nets [1, 27], Neural Nets [19, patterns, technological change and dif- are made simultaneously at known 16] and Bayesian Nets [2, 25]. Petri Nets ferent kinds of random perturbations; equilibrium prices. The known adap- are some of the most successful models and internal uncertainty due to the fact tive (tatonnement) processes, attempt- of Discrete Event Dynamic Systems and that each agent makes decisions with- ing to explain how disequilibrium are very good for representing asyn- out the full knowledge of states and prices move towards equilibrium, con- chronous distributed processes; however, actions of other agents. Economic verge only under strong assumptions. it is difficult to use them to represent ‘particles’ (agents, enterprises, coun- One of them is the reversibility of nontrivial agents with resource transfor- tries) do not follow strong laws like the decisions made at disequilibrium mation, exchange and control strategies. laws in mechanics and physics (for prices. This is a critical assumption Neural Nets are useful models for recog- example the law of gravity). They have for various imperfect and artificial nition of complex patterns which can be flexibility to choose different markets. The irreversibility of deci- present in multi-agent systems; however, behavioural patterns. Thus, both mod- sions leads to inefficiency of standard they lack tools for representing distri- els of uncertainty and behaviour of tatonnement procedures and thin mar- buted decision making. Bayesian Nets economic agents under uncertainty kets. It calls for approaches when the are good for processing incomplete infor- should be included in the system. equilibrium prices can be found with- mation, but again lack a structure for out making real decisions at disequi- agent representation. Therefore, a new • Complexity. Traditional economic librium prices. One of them is to create network like model is needed for repre- modelling deals with systems com- an artificial multi-agent market system, senting multi-agent systems.
Telektronikk 3/4.1998 35 − R = {rj | (ri,rj) ∈ P } ∈ + ∈ Another source of our insight comes i C2. For each aj A exists rk Oj such ∈ i from related work in computational econ- – the set of – resources to which oriented that rk Ii. omy and market-oriented programming, arcs point from resource ri. ∈ which resulted in the creation of several C3. For each rk Oi we have tools for distributed resource allocation The purpose of the Agent Net is to model + R ⊆ Ti ∪ Ii. in financial and other fields [14, 31, 32, the transformation of resources from R by k 33]. Still another contribution comes agents from A. Informally, R+is the set of i ∈ ∪ ∈ from the evolutionary economics, which resources which participate in the trans- C4. For each rj Ti Ii exists rk Oi ∈ + studies dynamic interactions between formation which results in resource ri. In such that rj Rk . economic agents [11, 23, 26]. order to transform resources agents make transactions between each other. The set Si(t) – state of the agent ai which is the In order to cope with uncertainties in- A+is composed of all agents who possess vector of quantities of resources from i ∪ ∪ ∪ herent in multi-agent systems we utilised resources needed in transformations per- Ti Ii Oi Ei. This vector is indexed approaches developed in stochastic pro- formed by agent ai. More formal de- by members of resource sets of agent ai gramming [5, 10, 13, 15, 20, 22, 24, 29]. finitions follow. and varies with time t. In this way the state S(t) of the whole multi-agent system 3 Agent Nets 3.1 Structure of agents is composed of the states of individual agents: S(t) = (S1(t), ..., Sn(t)). An Agent Net is composed of two ori- Agent Nets differ from other network F = {F jk (•), j ∈ O , k ∈ T ∪ I } – the set ented co-ordinated graphs like structures, notably Petri Nets, Neural i i i i i of transformation functions which define nets and Bayesian Nets, by a more in- {(A,N), (R,P)} (1) relations of the type volved node structure, which permits the where modelling of different types of agents y = F jk (x) i from real multi-agent systems. where y is the amount of internal or input (A,N) – the agent graph – is defined by resource r necessary to obtain amount x the set of na vertices A called agents and Each agent a is a tuple k i of output resource r . Similar transforma- the set of oriented arcs N ⊆ A × A. j (T , I , O , E , S , F , M , D ) (2) tion functions are defined for internal Agents possess an additional structure i i i i i i i i resources and they describe the amount which will be defined separately. Arcs where of input resources necessary to obtain the connect those agents which can be Ti ⊆ R – set of internal resources; these specified amount of internal resource. In involved potentially in transactions. resources are needed to model production this case such functions are used to de- capacities and technical capabilities; scribe processes of investment and de- (R,P) – the resource graph – is defined by velopment. the set of nr vertices R called resources and Ii ⊆ R – set of input resources; these the set of oriented arcs P ⊆ R × R. resources are obtained from agents M (t) – the information available to agent Quantity and other attributes may be i a at time t. Generally, it is some subset associated with each resource. belonging to A+in the process of i i of the state S(t) of the multi-agent system transactions; obtained with some delay and contami- For each agent a ∈ A let us define two i nated by errors. sets: Oi ⊆ R – set of output resources; these + resources are obtained by transformation D (t) – the set of strategies of agent a at A = {aj | (aj,ai) ∈ N} i i i from input resources with the help of time t. These strategies depend on the – the set of – agents from which oriented internal resources and constitute the offer state Si(t) and information Mi(t) and arcs point to agent ai; – of agent ai to agents from Ai ; include − Ai = {aj | (ai,aj) ∈ N} • Production strategies which define the Ei ⊆ R – set of exchange resources; – the set of – agents to which oriented amount of output resources to offer; these resources are exchanged by agent arcs point from agent ai; + ai with agents from Ai for input • Development strategies which define resources. the amount of internal resources to add Similarly for each resource r ∈ R we i to existing ones; define: These resource sets are connected with • Transaction strategies involving the + agent and resource graphs by the set of Ri = {rj | (rj,ri) ∈ P } selection of partner agents to make the constraints, in particular – the set of – resources from which offer of output resources to, fixing the oriented arcs point to resource r ; amount of exchange resource asked for i C1. For each r ∈ I exists a ∈ A+ such k i j i the output resource, selection of that r ∈ O . k j partner agents for obtaining input resources.
36 Telektronikk 3/4.1998 3.2 Evolution of Agent Nets Input resources. These are the resources competing voice services. In this case the used by agents for creating products and final resource would be ‘the need for The Agent Net evolves in continuous or services and satisfaction of needs. For voice communication’ measured, for discrete time. Its evolution is driven by example, in the case of the agent repre- instance, by distribution of time per day transactions between agents. Each agent senting an Internet provider, one of the for various types of customers. Input selects its production, development and input resources may be the lines which resources in this case may be ‘fixed transaction program according to the set he leases from a telephone company. In phone connection’, ‘mobile phone con- of its strategies and available informa- the case of the agent representing an nection’, ‘Internet connection’ and ‘other tion. Each transaction involves the Internet user, some of the input resources means’. exchange of input resources for exchange are a fixed local telephone service and resources, which change the volumes of the Internet connection. Input resources This resource structure is very flexible these resources in possession of agents. are bought by an agent in the market and and can easily be modified by reassign- The evolution of Agent Net is described may be stored. ing resources to different types and in more detail in [6]. aggregation/disaggregation according Output resources. These are the products to modelling needs. Agent Nets can be studied using graph and services into which agents transform theoretical methods and methods used input resources and which are offered to 4.2 Agents in the study of Discrete Event Systems. the market. For example, for an Internet There are many open research issues provider an output resource is the capac- Agents transform and exchange which are discussed together with some ity to provide an Internet connection of resources described previously. We properties of Agent Nets in [6]. given quality, while for a telephone com- developed generic agent structures which pany the output resource is the capacity can be specialised in the rich collection to provide a telephone connection. From of agents by specifying agent parameters 4 Informal discussion of these examples it is clear that the output for particular purposes. This structure resources for one agent are the input agent models permits us to model a variety of eco- resources for some other agents. Output nomic players from enterprises to indi- In this section we informally discuss the resources can constitute the offers to the vidual users. Such flexibility is important concepts described above and show how market and they can be stored. because we needed the capabilities to they can be translated in order to repre- model agents which combine the multi- sent economic reality. We again consider Internal resources. These are resources tude of industry roles. the main components of Agent Nets, but which are possessed by agents and are give them more informal treatment. necessary for transforming the input Roles. In the rapidly evolving informa- resources into output resources. Ex- tion economy one of the most important amples of such resources are qualified 4.1 Resources issues for a newly emerging company as manpower or production capacities. For well as for an established industry leader example, for an Internet provider his These are the elementary entities from is which industry roles to assume. Should Internet node would be his internal which the system is composed. Or, they the established fixed network provider go resource; for an Internet user it would be can be viewed as an alphabet in which into providing Internet services or form a his personal computer and specialised the system is described. In our termi- strategic alliance with a provider of cable software; for a telephone company it nology we consider resources to be any television? Thorough analysis of infor- would be her network. Input resources commodity or entity which is exchanged, mation industry roles can be found in [8]. can be expanded and otherwise developed satisfied, manufactured or in any other After preliminary analysis we understood and should be subjected to maintenance. way changed by economic agents rele- that all industry roles can be represented Money and input resources are needed vant to modelling purposes. Thus, be- in the alphabet of resources described for both maintenance and development. sides resources in the economic sense above, i.e. as transformation of a specific of the word, other examples of resources set of resources into another set of Final demand and needs. These are the are money, all kinds of products, services resources and their exchange. From this final resources which drive the economic and needs. In our system resources are resulted the fact that the agents them- activity of the system. They are not trans- divided into five types: exchange resour- selves can be represented in terms of formed or exchanged in the system and ces, input resources, output resources, this alphabet. constitute needs and demands of the end internal resources and final demand. user. What the final resources are very Thus, the generic agent structure in our much depend on the purpose of the mod- Exchange resources. The most common system consists of resource sets, trans- elling. Suppose, for example, that example of such resources is money, but formation functions and strategies. we model the penetration of the new other types can be modelled as well telecommunication voice service, like (information, rights, etc.). These are ob- Resource sets. There is a total set of voice over Internet. Then the final ligatory resources which are always pre- resources for all systems. Each agent is resource may be just ‘demand for voice sent in the system and whose flow is characterised by four subsets of this set, over Internet’ represented by some expert treated separately from the flow of other i.e. a set of input resources, a set of out- prediction. On the other hand, we might resources. This is due to their economic put resources, a set of internal resources be interested in looking more closely at function of exchange and because per- and a set of exchange resources. Input how this demand is formed according to formance of agents is often measured resources are all resources which are some hypotheses about behaviour of cus- in exchange balance terms. transformed by this particular agent into tomers and price and quality structures of
Telektronikk 3/4.1998 37 internal and output resources. For parti- ment strategies and purchasing services on the market. This agent is cular agents generated from the general strategies. All these strategies may in further characterised by the capability to structure some of these sets may be some cases be derived by solving substitute different products to satisfy empty. At each moment in time the state dynamic optimisation problems. In other the same need. For example, the need for of an agent is characterised by available cases such strategies can incorporate voice communication can be satisfied by exchange and internal resources and by principles of adaptivity and bounded fixed phone, mobile phone or voice over stocks of input and output resources. rationality. Internet.
Transformation functions. There are four Pricing strategies define the price which Pure supplier. This agent has only an sets of such functions in the general an agent offers for its output resources output resource which supplies for the agent structure: production functions, (products and services). In our case study price derived from the expert estimates. development functions, maintenance we implemented the principles of bound- This agent is useful for modelling the functions and satisfaction functions. ed rationality as follows. Each agent had supply of important products whose flow Production functions tell how much of a set of several strategies: keep the mar- we do not want to describe in much exchange, internal and input resources ket price, increase the price or decrease detail for modelling purposes. One ex- are needed for producing the given quan- the price based on previous history. At ample is the regulatory commission tity of the output resource. They have the each step an agent could choose from one which distributes frequencies for broad- following structure: of such strategies according to probabi- cast transmission. ψ lities which were updated according to vi = (a, v0) (3) their performance in terms of income and 4.3 Market where vi is the volume of specific input, revenue, similar to the theory of learning internal or exchange resource, v is the automata. 0 The market is the environment in which volume of the output resource and a are agents operate. At any time agents who production parameters. In the simplest Development strategies. If demands ex- produce output resources put their offers case these functions could be linear; ceed production capacities an agent can on the market. Each offer consists of however, we are specifically interested in choose between increasing the price or quantity and price of the specific the case of increased returns and expanding production capacities. De- resource. Agents who are customers for economies of scale. In such a case ψ(a,⋅) velopment strategies govern such ex- the input resources go to the market and is a concave function which may asymp- pansion taking into account the fact that choose between offers. For the case when totically tend to linear with increasing newly added capacities are becoming demand exceeds supply the system has argument. The simplest case of such a operational after some delay. the set of rules which sees to available function is the following: supply being distributed between cus- Purchasing strategies. These strategies 1+a2v0 tomers. Producing agents may then vi = a1v0 are employed by the agents for the selec- 1+a3v0 decide to increase the price for the next tion of offers for required input resources period and/or to expand production where the case describes increasing present in the market. The simplest capacities. There is a set of balancing returns when a > a and a < a corre- strategy is, of course, to choose the 2 3 2 3 mechanisms which are needed because sponds to diminishing returns. offer with lowest price. We take into the unsatisfied demand of one agent may account, however, that for real economic result in a decrease of its offer to the All other types of transformation func- agents the price considerations are not market which in turn may result in tions have the same structure (3) as pro- necessarily unique and allow customers diminishing satisfaction of demand of duction functions. Development functions to migrate between offers with different another agent. One possibility is to use describe amounts of input resources and prices with some dynamics dependent on the generalisations of Walras taton- exchange resources necessary for ex- other attributes of an offer. nement process [33]. panding production capacities for a given amount. Maintenance functions define At this moment we implemented some the amount of exchange and input basic set of strategies which is in the pro- 5 System MODAGENT resources necessary for maintenance of cess of expansion. internal resources and stocks of input and In this section we describe the architec- output resources. Satisfaction functions Specific agents are generated from this ture of the object oriented modelling sys- define the amount of exchange and input general agent structure by specifying its tem MODAGENT which implements the resources necessary for satisfying the elements. Here are some types of the concept of Agent Nets described in the need of end user. agents with which we experimented. previous two sections. It is conceived as a tool for modelling complex distributed Strategies. Strategies are actions which Production agent. This agent puts on the multi-agent systems discussed in Section 2. agents undertake in order to achieve spe- market products and services produced cific aims. Strategies depend on the from production capacities using input System MODAGENT shares the amount of exchange and other resources resources bought on the market, but does methodological approach with system available to an agent and on the informa- not have final needs to satisfy. These INFOGEN developed at the first phase of tion available on the states and strategies agents further differ by their set of strate- this project with support of Italtel, and of other agents. The general agent struc- gies. constitutes its further development. In [6] ture includes different types of strategies, one can find the description of INFO- for example pricing strategies, develop- End user agent. This agent satisfies the GEN and the first applications to the final needs by purchasing products and modelling of information economy.
38 Telektronikk 3/4.1998 MODAGENT, however, proved to be 5.1 General description of more than just a software tool: it consti- modelling choices in Market tutes a contribution in its own right to the MODAGENT modelling methodology of complex multi-agent systems. These will be illustrated in terms of our principal motivation: modelling of infor- Exchange Input Output Exchange Importance of object oriented approach. mation economy. Our approach, how- The complexity of the economic system ever, can be applied to a wide variety of under study is such that it is unrealistic to complex distributed systems. We con- Internal pursue the aim of creating a modelling sider economy to be made up by the system which would embrace all prob- following four components (see Figure 1): lems of interest. What is possible, how- • Resources which are the basic building Agent ever, is to create a generic modelling sys- block of an economy. They are under- tem whose structure would reflect the stood in very general terms and em- structure of the telecommunication econ- brace all ‘passive’ entities found in Figure 2 Relations between resource omy, and which would be endowed with economy: resources in the proper classes and classes Market and the capability of filling this structure with sense, products, services, information Agent particulars of the given problem. It is and derivatives, demand, consumer important that different components of needs. Resources can be transformed the system can be changed and specified from one to another, exchanged, con- without involving other components. The sumed, accumulated; object oriented approach provides the appropriate methodology for doing just • Agents which are the basic ‘active’ Market this, having the means to define classes entity of economy. They transform, which encapsulate specifics, hiding them produce, sell, consume, accumulate, from other parts of the system. offer, exchange resources, obtain Exchange Demand information on economy, formulate It also provides the means to develop and execute strategies in order to fulfil specific models from generic instance by their objectives; utilising the concept of class derivation. Need • End users who are a special kind of This makes it possible, for example, to agents, but which we decided to treat define the generic structure of the eco- separately for modelling purposes. EndUser nomic agent and derive from it specific They consume products and services agents with relatively little effort. In produced by agents in exchange for addition, object oriented approach en- money and possibly other exchange courages the developer to think in terms Figure 3 Relations between resource resources. In doing so they pursue the of the general structure of the application classes and classes Market aim of satisfaction of their needs; field which results in descriptive models and EndUser giving important feedback to both mathe- • Market facilitates and organises ex- matical modelling and applications. change between agents and between These considerations determined our agents and the end users. choice of object oriented paradigm for development of MODAGENT which is These main components constitute the user consuming Demand resources pro- implemented in C++. top level classes of the object oriented cured in the Market in exchange for hierarchy of MODAGENT. In what fol- exchange resources. Thus, resources lows we indicate the names of the classes belonging to the Input class of one like this. The object of class Economy agent may belong to the Output class is to organise interaction between objects of another agent and to the Demand of classes Agent, EndUser and Mar- class of the end user. The relations ket and simulate dynamics of this inter- between different resource classes and End Agents action in discrete time. classes Agent, EndUser and Market users are shown in Figures 2 and 3. Market Resources can belong to one of five top level classes: Input, Output, In- The complex relations which connect ternal, Demand and Need. Besides, agents and end users necessitated a fur- there are exchange resources which are ther structuring of Agent class into high handled by special class Accountant. level subclasses. Besides resource classes Input resources are procured in the Input, Output and Internal men- Market in exchange for exchange tioned above, these classes include: resources, are transformed by an agent Information handles information Resources in Output resources using Internal available to Agent; resources and are offered to the Market in exchange for exchange resources. Consumer procures and handles Figure 1 Top level structure of Economy Need resources are satisfied by the end Input resources;
Telektronikk 3/4.1998 39 Market, other agents and end users patterns of relations between industrial agents which may include simultaneous competition and collaboration. These patterns are further affected by different industrial strategies of agents resulting from incomplete information in their possession, different relative strengths Agent Manager and weaknesses. These experiments show that the system in its current state already permits us to perform modelling Information and qualitative analyses of these phe- nomena, explore dynamics of agent learning, describe complex relations of Seller Accountant competition and co-operation between providers of structural and infrastructural services, highlight roles of regulation, Output Supplier Producer Consumer Output and evaluate various market strategies.
6.1 Model of competition and collaboration between Internal providers of infrastructural and structural services
The economic system in question is Figure 4 Top level structure of class Agent made up from the following economic agents (see Figure 5).
1. End Users. They have need N1 which we here call ‘need for information ser- vices’ for whose satisfaction they have Producer transforms Input The structure of these classes will not some fixed renewable budget. This need resources into Output and handles be described here. The top level structure can be satisfied by information services expansion of Internal resources; of Agent is shown in Figure 4. Many S1, S2 and S3. agents will not possess fully developed Supplier handles and dispatches components of this structure; in such Service S describes traditional informa- Output resources produced by 1 cases some components will be empty. tion services which require infrastruc- Agent; tural services of network operators and Seller markets Output resources; Detailed description of the MODAGENT are provided by network operators them- structure can be found in [18]. The next selves, like obtaining any kind of infor- Accountant handles the flow of section is dedicated to a case study which mation by traditional telephony, yellow exchange resources, keeps balance refers to dynamics of relations between pages, etc. and generally keeps track of Agent providers of basic telecommunication performance; services and providers of information Service S2 refers to a new kind of infor- Manager defines policies executed services which are built on top of mation services which also require by other members of Agent. telecommunication networks. infrastructural services of the network operator, but which can be provided by both network operators and independent 6 Numerical experiments companies which lease lines from net- with MODAGENT: work operators. Think, for example, End users Agents about Internet based provision of infor- dynamics of relations mation services. S 1 between providers A Service S3 refers to information services S 1 2 I1 of structural and infra- relatively independent of infrastructural services of network operators, like, for S2 structural services A example, radio and TV services. N1 2 This section is dedicated to MOD- S These services are partially substitutable 3 AGENT modelling of industrial relations A3 between each other. End users choose in the emerging information industry. In between different services according to particular, we model the combination of the following demand generation model. structural and infrastructural roles by The attitude of each end user towards any Figure 5 Providers of infrastructural and different industrial players. This combi- particular service S is characterised by structural services nation may result in complex dynamic i
40 Telektronikk 3/4.1998 the service utility function fi(x) which is a information services S1 and S2 and Providers choose between these three non-decreasing function varying from 0 infrastructural service I1 necessary for strategy classes according to information to 1 and defines the fraction of need N1 provision of S1 and S2. Alternatively, we about the market behaviour. If profit satisfied by amount x of service Si. This refer to this agent as Network Operator. increased during the current period then function is characterised by two para- We assume that provision of service S2 the current strategy is maintained or meters: besides requiring infrastructural service enforced; otherwise the strategy is likely I1 also requires internal capability C1 to change. This is done according to the ai – the maximal possible fraction of which needs to be developed. All other learning process which uses information need which can be satisfied by a given resources necessary for provision of ser- about market behaviour and is organised service when its amount $x$ tends to in- vices S1 and S2 are expressed in terms of similarly to the learning process in the finity, we assume that all ai sum up to 1. money. theory of learning automata. That is, each strategy is characterised by probability bi – the incremental fraction of the need 2b. Provider of structural services A2. according to which this strategy is satisfaction by service Si for small values He provides information service S2 and chosen in the next step. These proba- of x. The end user chooses between ser- needs to buy infrastructural service I1 bilities are updated each period: the vices by maximising his need satisfaction from A1. Alternatively, we refer to this probability of strategy which improves within the given budget (perhaps with provider as ‘Information Service Pro- some specified performance criterion is some error). This permits us to model a vider’. Thus, he is a competitor to A1, al- increased while probabilities of other non-homogeneous user population with though in order to compete he needs the strategies are decreased. However, the following service adoption cycle. service I1 provided by A1. Take, for probabilities of all strategies remain posi- Suppose that there is some established example, as A2 an Internet provider who tive. This approach permits us to model a service and some emerging service. In leases lines from a network operator to rich set of possible behaviours. the early stages of service introduction provide information services, and take as there will be some fraction of ‘early bird’ A1 a network operator who is also en- We took the current profit as the agent users which will adopt a new service gaged in providing information services. performance measure. The approach just even if it costs substantially more than Think, for example, of AT&T or ‘Baby described permits us to reflect the fact traditional services, it is enough that the Bells’ from one side and America that the profit maximisation is the most new service has some new appealing OnLine and CompuServe from the other important, but not unique driving force feature. In the later stages the main body side. of agent behaviour. Therefore, we also of users will adopt the new service, but permit keeping the strategy which led to only if it becomes price competitive with 2c. Provider of structural services A3. He profit decrease, but with smaller proba- the old service. And finally, there will be provides information service S3 and is bility compared to the case when this a fraction of users who will switch to the independent of both A1 and A2. Think, strategy increased the profit. new service only if it becomes substan- for example, of a newspaper publisher or tially cheaper than the old service. an owner of TV channels. His economic We considered the cases when provider function is to provide a service which can A1 varies his offer of both infrastructural Normally, there is more than one pro- substitute to some extent both S2 and S1 service I1 and structural service S2 keep- vider of the same service on the market in case they become too expensive or ing fixed the offer of structural service who offer services with different prices. inadequate in some other respect. We S1, while provider A2 varies the offer of We assume, however, that there are more assume for simplicity that he provides service S2. attributes to a service than the price; ser- service S3 for a fixed price. vices from different providers may differ Our objective was to model dynamics in many other respects, like quality, etc. 3. The objectives of economic agents. of competition and co-operation between Thus, not all end users select the provider We assume that end users maximise the providers A1 and A2, penetration of with the cheapest service. However, the satisfaction of their needs for information service S2 and dynamics of total market price remains an important attribute of a services as described above. Providers for information services S1 and S2 under service and market quotas of different maximise their profit by choosing among different scenarios about: producers increase and decrease depend- different strategies. These strategies • Market and expansion strategies of ing on price. This permits us to differen- belong to three classes: both providers; tiate between basic services with stan- • Increase market offer, decrease price dard quality and more expensive services • Different provision costs for service S ; and expand provision capacities if nec- 2 with enhanced quality. essary; • Different policies of market regulation. Finally, we take into account one more • Decrease market offer and increase characteristic of end user behaviour: even price, maybe moving to a more 6.2 Description of scenarios if services are identical end users follow specialised market niche; with some delay the relative price Each scenario is characterised by the • Keep the price and market offer con- dynamics of different producers. following components: stant. 2. Producers. There are three groups of 1. Relative production costs for provision Different strategies from the same class producers in our economy: of service S by providers A and A . differ quantitatively, for example by the 2 1 2 Here we consider two alternatives: costs amount of price decrease or increase. 2a. Provider of structural and infrastruc- of provider A1 (‘Network Operator’) are tural services A1. He provides structural considerably higher then the costs of
Telektronikk 3/4.1998 41 Scenario Provision costs Policies Regulation Maintenance small variation then we say that he pur- costs sues ‘weak’ policies. Obviously, weak- ness or strength is relative between A A high, A low A weak, A strong No Low providers. The strength of the provider is 1 2 1 2 measured numerically by the maximal possible price changes during a certain B A1 high, A2 low A1 strong, A2 strong No Low period of time. C A1 high, A2 low A1 strong, A2 weak No Low 3. Market regulation which in our case D A1 high, A2 low A1 strong, A2 weaker than in C No Low consists in fixing allowable price range for services by outside authority or by E A1 high, A2 low A1 strong, A2 weaker than in D No Low agreement between providers. We con- sider the following cases: F A1 high, A2 low A1 strong, A2 weak as in E Bounded price Low of S2 for A1 • Absence of market regulation; • Upper bound is fixed on the price of G A1 high, A2 low A1 strong, A2 weak as in E Low maximal Low infrastructural service I1; price for I1 • Lower bound is fixed on the price of H A high, A low A strong, A weak as in E High maximal Low 1 2 1 2 structural service S2; price for I1 • Both upper bound on price of I1 and I A1 same as A2 A1 weak, A2 strong No Low lower bound on price of S2 are fixed.
J A1 same as A2 A1 strong, A2 strong No Low 4. Costs for maintaining provision capa- bility of service S2. We wanted to con- K A1 same as A2 A1 strong, A2 weak No Low sider maintenance costs separately from production costs because in some cases L A1 same as A2 A1 strong, A2 weaker than in K, No Low service providers tend to overdimension same as E provision capabilities in order to increase competition flexibility. We considered M A1 same as A2 A1 strong, A2 weak as in E High maximal Low two cases: price for I1 • Low maintenance costs; N A same as A A strong, A weak as in E Low maximal Low 1 2 1 2 • High maintenance costs. price for I1, half as in M The model permits us to study also the cases with different economies of scale O A1 same as A2 A1 strong, A2 weak as in E Minimal price Low for different providers, different end user for S2 preferences and different initial market P A1 same as A2 A1 strong, A2 strong Minimal price Low conditions which we do not consider in for S2, maximal this paper, however. Each individual price for I1 scenario can be obtained by choosing one of the scenario components described Q A1 same as A2 A1 strong, A2 strong No High above. We considered 19 scenarios in- dexed by letters A–S which are summar- R A1 same as A2 A1 strong, A2 strong Minimal price High ised in Figure 6. for S2 For each of these scenarios we simulated S A1 same as A2 A1 strong, A2 strong Minimal price High the economic system described above for S2, maximal starting from the same initial conditions: price for I1 initial provision capacities for service S2 and initial prices for service S2 which Figure 6 Summary of scenarios were equal for both providers. Results of simulations contain evolution of the following quantities: 1 Profit of both providers;
provider A2 (‘Information Service decrease, possible capacity expansion 2 Revenue of both providers; Provider’) and the case when their costs and capability to add value to a service 3 Price for information service S for are roughly equal. in case of the price increase. If the pro- 2 both providers; vider possesses the capability (or willing- 2. Capability of the provider to change ness) to vary his market offer consider- 4 Volume of service provision for ser- the market offer from one period to an- ably we say that he pursues ‘strong’ poli- vice S2 for both providers; other. This involves the price increase or cies, if he is capable only of relatively
42 Telektronikk 3/4.1998 5 Provision capacities for service S2 for narios described above. For the main part both providers; we are here going to analyse the evolu- tion of profit, since in this experiment 6 Revenue breakdown for A $ between 1 the profit maximisation was taken to be services S , S and I ; 1 2 1 the ultimate goal of providers. 7 Total market value for information ser- vices S1 and S2 with market shares of 1. Group of scenarios when Network providers A1 and A2; Operator A1 has higher provision costs for service S2 compared to Information 8 Price for infrastructural service I1. Service Provider A2 (scenarios A–H)
Evolution of these quantities are shown These scenarios cover cases when A1 Figure 1.A in figures named ‘Figure X.Y’ where X does not have sufficient knowhow to pro- indicates the number of quantity as given vide the competitive service S2 for differ- above and varying from 1 to 7; Y varies ent economic and organisational reasons. from A to S indicating scenarios defined Figures 1.A, 1.B and 1.C illustrate the above. Due to the lack of space we give point of importance of strong policies. here only a part of these figures. By strong policies we here mean the capability of the provider to rapidly 6.3 Analysis of results change the price of the offered service and, consequently, meet demand varia- tions. Since, in the case of price decrease, Results exhibit many interesting phe- demand may increase considerably this nomena which have convincing eco- also means the capability of the service nomic interpretations, in particular provider to make sufficient investment in • Benefits of market regulation which order to meet increased demand. On the may be performed either by appropri- contrary, we are speaking about weak ate regulatory body or by agreement policies when a service provider varies between providers. Such regulated the price and quantity of offered service competition permits providers to sur- by relatively small steps. Figure 1.B vive who offer superior services but cannot expand the offer rapidly, in Figure 1.A shows that in the case when many cases safeguarding the total mar- provider A1 employs weak policies and ket value and in many cases being ben- provider A2 employs strong policies the eficial to all players since it stabilises profit share of A2 gradually increases the market; while the profit share of A1 gradually decreases. Profits of A increase com- • Tendency of Network Operator to con- 1 pared to profits of A when both pro- centrate on his core business when his 2 viders use the strong policies (Figure information service provision costs are 1.B). The profit share of A increases high and his tendency to compete in 1 even more when he uses the strong poli- information services when his costs are cies while A uses the weak policies (Fig- comparable with Information Service 2 ure 1.C). In all cases the system reaches Provider; equilibrium state after some time and • ‘Strong wins’: importance of pursuing continues to oscillate around equilibrium aggressive market and expansion poli- state. The equilibrium profit distribution, cies (for those who can afford them), however, depends dramatically on the Figure 1.C the provider with strong policies relative strength of policies employed by assures himself the largest profit share providers: the stronger they get, the even with somewhat inferior provision larger the share. Analysis of volumes and costs; different revenue items of provider A1 shows that after the initial attempt to • High instability of non-regulated mar- compete with A on the provision of ser- ket when everybody pursues strong 2 vice S he decides to concentrate on his policies, price wars; 2 core business, provision of infrastructural • Emergence of qualitatively different service I1, and gradually pulls out of pro- dynamical patterns of the system viding S2; the faster he does it, the evolution which in some cases may stronger his policies are (Figure 6.B). include oscillations in vicinities of dif- This is understandable, given that his ferent equilibria and in other cases provision costs are higher than for markedly chaotic behaviour. provider A2. In this way provider A2 obtains the major part of the market for Let us consider in more detail some of information services S1 and S2, while the simulation results for different sce- provider A1 retains only a small part of Figure 6.B
Telektronikk 3/4.1998 43 the market related to traditional service S1 (Figure 7.B).
What happens if pro- vider A2 is only capable of using even weaker policies? In this case, starting from some point, provider A1 is Figure 7.B capable of driving pro- Figure 1.F vider A2 into bankruptcy by increasing the price for service I1 (Figure 1.E). This has unfortu- nate consequences even for provider A1 since he loses his revenue from providing infrastructural service I1 and is not capable of substituting the lost profit by provid- ing service S2 himself. He is capable of partially recovering the lost rev- enue by increasing pro- vision of service S1 (Fig- Figure 1.E ure 6.E), but many users Figure 7.F migrate to service S3 offered by provider A3. Consequently, both the profit of A1 and the total market for services S1 and S2 decrease sub- stantially (Figure 7.E).
Let us now take a look at the effects of regulation in this case. The first possibility is to force provider A1 to maintain the capability for provi- sion of service S2. This somewhat preserves the market for S even in Figure 6.E 2 Figure 7.G case of the bankruptcy of A2, but both the profit of A1 (Figure 1.F) and the total market for in- formation services S1 and S2 (Figure 7.F) de- crease substantially, al- though not to the same extent as in the absence of regulation.
Another possibility of regulation is to impose the upper bound on the price of infrastructural service I1. This permits provider A2 to survive even in the case of weak Figure 7.E Figure 1.G policies. This brings
44 Telektronikk 3/4.1998 benefit also to provider A1 and increases the total market for informa- tion services (Figure 7.G) since the system reaches equilibrium with larger profits for A1 (Figure 1.G) than in the case of bankruptcy of A2. Figure 1.H shows a Figure 1.H borderline situation Figure 6.J where the upper bound for the price of I1 is higher than in the case depicted in Figure 1.G. Provider A2 starts to lose money, but finally man- ages to recover.
Note that this price re- straint can be imposed by the regulator body, but it can also be a result of self-restraint of pro- vider A1.
2. Group of scenarios Figure 1.I when Network Operator Figure 7.J A1 has similar provision costs for service S2 com- pared to Information Service Provider A2 (scenarios I–P) These scenarios cover cases when A1 is capable of competing with A2 even on service S2. Fig- ures 1.I, 1.J and 1.K are similar to Figures 1.A, 1.B and 1.C respectively, and show that in this case profit distribution is shifted substantially in favour of provider A1, Figure 1.J since in this case he gets Figure 3.J revenue both from pro- vision of S2 and from provision of the infra- structural service I1. Similar to the previous case, the equilibrium profit distribution criti- cally depends on the rel- ative strength of policies which can permit providers: the stronger policies the provider pursues, the larger prof- its he gets if the policies of other providers are fixed. To see this, com- pare Figure 1.I where Figure 1.Kprovider A1 employs Figure 1.J.A
Telektronikk 3/4.1998 45 weak policies and pro- vider A2 employs strong policies, with Figure 1.J where both providers pursue the strong poli- cies, and with Figure 1.K where A1 uses the strong policies while A2 uses the weak policies.
Figure 3.J.AThe case when both Figure 3.P providers pursue strong policies merits further discussion. If we look at the revenue breakdown for provider A1 (Figure 6.J) and at behaviour of market shares (Figure 7.J) we see that the sys- tem is very unstable and exhibits chaotic beha- viour. Price dynamics show that providers are engaged in a price war (Figure 3.J). Chaotic behaviour can further be seen in Figures 1.J.A, 3.J.A and 7.J.A which Figure 7.J.A Figure 1.L differ from Figures 1.J, 3.J and 7.J by a con- siderably greater time span. In this case regula- tion (or mutual agree- ment) can have a strong stabilising function. The introduction of lower bound on service S2 con- siderably reduces the profit oscillations (com- pare Figure 1.O with Figure 1.J.A). Such oscillations are reduced even more if in addition the upper bound on the price of infrastructural Figure 1.P Figure 7.L service I1 is introduced (see Figure 1.P). Chaotic oscillations in market shares and prices are also reduced consider- ably (compare Figure 7.J.A and 7.P, Figures 3.J.A and 3.P). Figure 3.P shows another inter- esting consequence of price restraint: provider A2 concentrates on the provision of higher qual- ity services at a some- what higher price, while provider A1 concentrates on provision of basic services at lower price. Figure 7.P Figure 6.L
46 Telektronikk 3/4.1998 Similar to the previous case, provider References A1 succeeds in driving A2 out of business if the policies of the latter are too weak 1 Archetti, F, Gaivoronski, A, Scio- (Figure 1.L). This, however, does not machen, A. Sensitivity analysis and lead neither to the profit decrease for A1 optimisation of stochastic Petri nets. nor to the decrease of the total market Discrete Event Dynamic Systems : value for information services (Figure Theory and Applications, 3, 5–37, 7.L). This is because in this case provider 1993. A1 can provide the competitive service S2 and takes over the customers of provider 2 Archetti, F, Gaivoronski, A A, Stella, A2 (Figure 6.L). Again, regulation which F. Stochastic optimisation on Figure 1.M fixes the upper limit for the price of I1 Bayesian Nets. European Journal of permits provider A2 to survive (Figure Operations Research, 101, 360–373, 1.M) but this does not have any bene- 1997. ficial effect on the market contrary to the previous case. 3 Arthur, W B. Bounded rationality and inductive behavior (the El Farol Predictably, higher prices for maintaining problem). American Economic provision capabilities lead to lower prof- Review, 84, 406–411, 1994. its, but do not change the qualitative pic- ture (Figure 1.Q without regulation and 4 Arthur, W B. Increasing returns and Figure 1.S with regulation). path dependence in the economy. Ann Arbor, Michigan University 7 Summary Press, 1994. 5 Birge, J R, Wets, R J-B. Computing We have here presented a general Agent bounds for stochastic programming Net methodology for modelling complex problems by means of a generalized distributed multi-agent systems found in Figure 1.O moment problem. Mathematics of telecommunications, and presented a Operations Research, 12, 149–162, prototype of the simulator MODAGENT 1987. for simulation of such systems. Our ref- erence point was its application to 6 Bonatti, M, Ermoliev, Y M, Gaivo- modelling of information industry, al- ronski, A A. Modeling of multiagent though it is applicable also to other systems in the presence of uncer- multi-agent systems. tainty : the case of information econ- omy. Laxenburg, IIASA, 1996. This project is in the stage of further (Technical Report WP-96-94.) (To development. Some of the issues which appear in Journal of Robotics and we address now are the following: Autonomous Systems, 1998.) • As we have seen multi-agent systems exhibit widely different dynamics 7 Bonatti, M, Gaivoronski, A A. Eco- under different values of system nomic modelling for strategical plan- parameters. The tools are necessary ning and top level engineering of which would permit us to identify the information infrastructures : overall regions of stability of certain equilib- approach. In: Proceedings of NET- Figure 1.Q rium points and regions where one WORKS’96, Sydney, 1996. strategy is superior with respect to another strategy. One possibility to 8 Bonatti, M et al. Information industry develop such tools is to extend to the enterprize model, reference models multi-agent systems the theory of sen- and items to be standardized and sitivity analysis developed for Discrete how. European Telecommunications Event Dynamic Systems [1, 9, 17, 21, Standards Institute, March 1996. 28, 30]. (Report ETSI/EPIISG(96)16.) • Development of robustness notions for 9 Cassandras, C G. Discrete event sys- the design of enterprise strategies and tems : modeling and performance their evaluation in a multi-player en- analysis. Irwin Publishers, 1993. vironment. • Game theoretical approaches for strat- 10 Dempster, M (ed.). Stochastic pro- egy evaluation. gramming. London, Academic Press, 1980. Figure 1.S
Telektronikk 3/4.1998 47 11 Dosi, G, Nelson, R. Evolutionary 19 Hertz, J, Krogh, A, Palmer, R G. discrete event processes. Annals of theories in economics : assessment Introduction to the theory of neural Operations Research, 39, 173–194, and prospects. Laxenburg, IIASA, computation. Redwood City, Addi- 1992. 1993. (Technical Report WP-93- son-Wesley, 1991. 064.) 29 Plambeck, E L et al. Optimizing per- 20 Higle, J L, Sen, S. Stochastic de- formance functions in stochastic sys- 12 Ermoliev, Y, Gaivoronski, A A. composition : an algorithm for stage tems. Mathematical Programming, Stochastic quasigradient methods for linear programs with recourse. Series B, Approximation and Com- optimization of discrete event sys- Mathematics of Operations Re- putation in Stochastic Programming: tems. Annals of Operations Re- search, 16, 650–669, 1991. State of the Art, 1993, 1993. search, 39, 1–39, 1992. 21 Ho, Y C, Cao, X R. Perturbation 30 Rubinstein, R Y, Shapiro, A. Dis- 13 Ermoliev, Y, Wets, R J-B (eds.). analysis of discrete event dynamic crete event systems : sensitivity anal- Numerical techniques for stochastic systems. Boston, Kluwer, 1991. ysis and stochastic optimization via optimization. Berlin, Springer, 1988. the score function method. Wiley, 22 Kall, P, Wallace, S. Stochastic pro- 1993. 14 Even, R, Mishra, B. CAFE : a com- gramming. New York, Wiley, 1994. plex adaptive financial environment. 31 Steiglitz, K, Honig, M M, Cohen, New York, Courant Institute of 23 Lane, D A. Artificial worlds and L M. A computational market model Mathematical Sciences, 1996. (Tech- economies. Journal of Evolutionary based on individual action. In: Clear- nical report.) Economics, 3(2 and 3), 89–108 and water, S (ed.). Market-based control 177–197, 1993. : a paradigm for distributed resource 15 Gaivoronski, A A. Approximation allocation. Hong Kong, World Scien- methods of solution of stochastic 24 Mulvey, J M, Ruszczynski, A. A new tific, 1995. programming problems. Cybernetics, scenario decomposition method for 18, 97–103, 1982. large-scale stochastic optimization. 32 Waldspurger, C A et al. Spawn : a Princeton Unversity, 1992. (Techni- distributed computational economy. 16 Gaivoronski, A A. Convergence cal Report SOR-91-19.) IEEE Transactions on Software En- properties of backpropagation for gineering, 18, (2), 103–117, 1992. neural nets via theory of stochastic 25 Neapolitan, R E. Probabilistic gradient methods. Part I. Optimiza- reasoning in expert systems. Wiley, 33 Wellman, M P. A market oriented tion Methods and Software, 1994. 1990. programming environment and its application to distributed multi-com- 17 Gaivoronski, A A, Shi, L Y, Sreeni- 26 Nelson, R, Winter, S. An evolution- modity flow problem. Journal of vas, R S. Augmented infinitesimal ary theory of economic change. Cam- Artificial Intelligence Research, 1, perturbation analysis : an alternate bridge, MA, Belknap Press of Har- 1–23, 1993. explanation. Discrete Event Dynamic ward University Press, 1982. Systems : Theory and Applications, 2, 121–138, 1992. 27 Peterson, J L. Petri net theory and the modeling of systems. Prentice- 18 Gaivoronski, A A. MODAGENT, Hall, 1981. users manual. Trondheim, Norwe- gian University of Science and Tech- 28 Pflug, G. Gradient estimates for the nology, 1998. (Technical report.) performance of Markov chains and
Alexei A. Gaivoronski (45) is Professor of Operations Research at NTNU, Trondheim. He has an M.Sc. from Moscow Institute for Physics and Technology and a Ph.D. from V. Glushkov Institute of Cybernetics, Kiev, both in applied mathematics specialising in operations research and optimisation. He has worked in both academia and industry, and his scientific interests are optimisation, mod- elling of stochastic systems, manufacturing and telecom- munication models. He has published more than 50 papers on these topics in international journals and books. e-mail: [email protected]
48 Telektronikk 3/4.1998 Technological Evolution: Some Strategic Considerations
JAN A. AUDESTAD
1 Introduction Plain telephony is an example of a cash category 1) to category 2); that is, be- cow. In developed countries the market is come products with low margins and lit- The strategic problems we are facing saturated and very little growth is ex- tle or no growth. From a business point today in telecommunications are all pected. This market is, however, difficult of view, such products should be termi- owing to the enormous growth of com- to assess for several reasons: nated. These products may then become plexity [1]. The complexity is present in a real problem for the operator: the prod- • The telephone network carries services all aspects of telecommunications. Below ucts must be retained for societal reasons, which are not just plain telephony; we will give an analysis of the technolog- and the products will still represent a examples are telefax and Internet ser- ical evolution and its consequences on major part of the overall telecommunica- vices (at least in local areas), and markets and research: how it is now and tions market, perhaps 70 % or more. The transaction services which also use the how we expect it will change during the size market does not matter as long as telephone network as carrier; next few years. The analysis will com- there is nothing to earn in it. prise the existing products, the new con- • Some of these services are fading out tent market, sensor technology and, such as telefax, since telefax is re- If the company has other products in cat- finally, nanotechnology. placed by e-mail; egory 2) without such constraints, they should be terminated before they cause • Some of the services using the tele- Some strategic problems in this context losses to the company unless they create phone network are increasing such as are a secondary stream of income. This is an Internet services; important consideration for all the prod- • How much resources should we allo- • Services which were supplementary, ucts. cate to further evolution of traditional are now substituting each other – products? mobile telephony is a good example Products in category 3) are particularly • Why do we not earn money on the new here; dangerous because they may indicate the content rich services? possibility of wrongly priced products or • Growth in mobile telephony causes that the cost is too high for these prod- • Will sensor technology, nomadic com- secondary growth in telephony ucts. There are many indications that sev- puting and wearable computing create because one of the persons connected eral of the Internet products are in this new opportunities for telecommunica- is generally located in the fixed net- category where we by Internet products tions? work. mean all the individual products for pro- • If nanotechnology ever materialises, cessing, storing, marketing, billing, From this it is not easy to interpret the will it have any impact on telecommu- maintaining and operating content rich evolution of the market for telephony by nications? services. Two related problems are that only measuring the traffic on the net- the income may come through different work; we must also differentiate between market channels (for example, creating the different usages of the network. The 2 Traditional Products more demand for ISDN installations), reason for this is that, by only measuring and that this cross-subsidisation may be [2] contains one way of assessing the the traffic, it is not evident how money against regulation. In more extreme cases portfolio of products. The method con- should be used for investments and re- the revenues may come via apparently sists in comparing each product in terms search: investments in plain telephony unrelated channels such as sales of of four variables: technology or in IP networks, or research energy. Telecommunications may then in basic switching technology or in new • Current margin created by the product; be delivered at very low prices because types of sources such as sensors? the infrastructure is also used for remote • Expected short term growth in revenue; control of the energy delivery. One likely option here is to invest as lit- • Current total revenue created by the tle as possible both in money and in- product in the market; Mobile communication has been a prod- tellect on plain telephony, and rather uct of category 4). Because of oligopolic • The market share the company has of direct the efforts towards the new oppor- competition with focus on price wars and this revenue. tunities. As I will explain in Chapter 3, it new roaming conditions for operators does not really matter on which network without their own networks, this product Products may then be viewed as belong- new services are transferred as long as may move towards category 3); that is, ing to four categories [9]: the service meets the expectations of the maintaining a high growth while the mar- market. 1 ‘Cash cows’ or products with high gin is going down. margins but with small or no growth Plain telephony is a cash cow now but potential; The strategic problems are then: competition, and in particular regulation, 2 Products with low margins and little or may lower the prices and reduce the mar- • To decompose products into parts no growth potential; gins unless the cost is reduced. This may which may be products on their own require a novel and innovative way of – such as in the example of telephony 3 Products with low margins and much allocating costs between different usages explained above – and to analyse each expected growth; of the network. such part separately; 4 Products with high margins and much • To identify to which categories the expected growth. Plain telephony and infrastructure different products of the portfolio products like access and transport may belong; – because of price pressure – move from
Telektronikk 3/4.1998 49 • To evaluate how competition and regu- Independent lation may move products from one companies category to another, and to assess the impact of such changes; • To assess the impact of products of category 2) if they are required to be retained by society.
3 Content Rich Services access Many content rich services are part of the inter- Dedicated existing portfolio of telecommunications ISDN ATM connect IP network operators. However, as I see it, the understanding of such services is much related to the problem of dividing the main product into parts. This may be done as shown in Figure 1. In the exist- ing context the content rich services are access related to the Internet. This may change; IP supporting Independent networks companies
Figure 2 Technical arrangement
Appl. Appl.
however, the general concept of Figure 1 connected to the networks via access sys- Web is likely to persist. tems. The servers communicate by using Appl. Appl. the Internet Protocol (IP). For simplicity In order to understand the services we I have not shown additional equipment divide them into three separate entities: such as routers required for supporting the web, the Internet servers, and the this communication. The operation and Browser Browser supporting network. These three layers management of servers are businesses are independent in the sense that the independent of network operation. Like technology may be replaced independ- networks, there may be a large number of ently and at different rates in the three companies in the business of offering layers. The different layers also represent servers (and routers and other facilities). separate services or sets of services. What is important is that ownership, Servers location and operation of servers are The Internet uses several types of net- independent of the supporting network works for transporting bits: telephone infrastructure. The technical arrangement networks, ATM networks, mobile net- is shown in Figure 2. This architecture works, cable TV networks, and so on. will be stable for a long time: the cost of There may be several networks of the replacing the network infrastructure is same or different type in tandem. The too high. This also applies if dedicated only requirement is that these networks IP networks are introduced. interwork with one another without destroying or changing the bits trans- The application (or web pages or soft- ferred from one end to the other. The ware programs) are shown as a separate Network terminals are connected via some type plane. This is so because the applications of access to the network. There is no are separate businesses: electronic news- requirement concerning the type of ter- papers, electronic trade, information ser- minal to be used except that it must sup- vices, education, office support, and so port the browser software which is the on. These companies own the application ‘terminal’ as seen at the application level. but they may outsource the processing and the management of the application to The networks also interconnect the companies owning servers, or they may servers which are computers designed for own their own servers. In the latter case running multiple applications in a dis- they are present in two businesses at the Figure 1 Model for Internet tributed environment. The servers are same time: applications and servers. The
50 Telektronikk 3/4.1998 Note also that content providers may co- ance with your habits and interests Content provider operate in delivering a joint service. learned by earlier visits to the mall). Content One problem with the Internet is that it is It is also difficult to enter many of the Appl. difficult to earn money on it. There may content businesses. One reason is that be several reasons for this: they may require co-operation com- pletely different from other systems. • The co-operation between hosts and This implies also ownership of the busi- Processing network operators may be such that the ness. In [1] the C4 convergence was host does not receive enough money explained. In the early phases of this from the network operator in order to development it was assumed that new Host be adequately paid for the value companies or at least firm alliances Processing created; User would form along the C4 axis content, • It may be the other way round: the computers, communications and con- price paid by the host for using the sumer electronics. So has not happened network may be too small; [3]. The reasons seem to be that such ver- tical integration is not economically opti- • The use of the network may be too Transport mal for each of the actors along the axis, expensive for the host and the content and that the need for new co-operation provider so that they lose money on changes faster than the time required to the service; Network operator build and stabilise a new company or Access • There may be much arbitrage where alliance. In the most complex cases com- the content provider and the host use panies may co-operate in one area and networks paid for by others, such as compete in others. governments; the ‘old’ Internet con- sisted of mainly ‘free’ circuits; This type of dynamic market, compe- tition and co-operation has not existed Figure 3 Customer relationships • The market mechanisms are really not before and is new to the participants in understood: the Internet may be a ser- the market. Traditional administration vice offer to people without money but and value monitoring does not apply to creating much communications which such systems. I believe that this is the means that prices accepted by the main reason why the content market has market are low. not materialised. Another problem with content is that the model is not changed by this or if some service may simply not be good enough. company owns all three layers. 4 Sensors, Wearables and Take electronic trade as an example. If you know exactly what you want, it is Nomadic Computing At the application level addressing is simple. If you do not, it is not so easy. done by hypertext pointers. The server The Internet as a commercial product is On one screen we can only show a small maps the hypertext pointer to an IP only five years old. Sensor technology, fraction of the goods you may see by address which is again mapped to tele- as we shall describe it here, is even strolling through a shop. Take the phone numbers, mobile subscriber num- younger. Of products now coming are “world’s largest bookstore” – the Ama- bers or whatever numbering system is video cameras, radars and navigation re- zon database. They claim to have two used by the network. ceivers on one chip. The video camera million titles, which is three times more is complete with lens and radio commu- than the biggest physical bookstore. If Generally, we may then have a structure nication. The navigation sensor can be you are just going to browse through as shown in Figure 3. The user is a cus- sewn into clothes and utilise the GPS such a big store, it will take time. tomer of the network operator (access). standard for determining position. The Suppose that you can read one book In addition the user may be a customer of radar can be used for automatic measure- title and synopsis every minute, then the host (owner of servers) and the con- ment of the oil level in motors and the the browsing takes 30,000 hours. If you tent provider. The content provider is a content of tanks. spend 12 hours every day on this en- customer of the host, and the host is a deavour, it will take you six years to fin- customer of the network operator. The artificial nose is under development. ish the browsing. Then you may just as Still it is not at microchip size but that well start again because many of the The payment for using the applications will come soon. Already gass chro- titles are either obsolete or new! may be made to the network provider matographs are down to this size with the who pays the host and directly or in- channel etched into the semiconductor This problem of electronic trade is purely directly the content provider; or the cus- which also contains the sensing device, psychological and not technical. It is easy tomer pays the content provider who the spectral analyser and the output for- to make the technology: virtual reality pays the other operators. The customer matting device. Microprocessors may shopping malls, huge databases like the may also pay all three independently. replace the bar code on merchandise. The Amazon, and customer specific adver- The payment method may depend on the electronic handcuff monitors people sen- tisements and construction of the mall application. tenced to mild imprisonment with some (where shops may be shown in accord- restrictions on their mobility. The list of
Telektronikk 3/4.1998 51 possible devices is increasing day by day Some of these technologies are not new; bler devices [6], [7]. Some well-known [4]. they are only reappearing in a new form results are producing the IBM logo at taking advantage of the miniaturisation. atomic level made from 35 atoms of Some of these devices are wearables Early systems were tracking devices on xenon; AT&T’s ‘three little gears’; and and wearable computers; that is, devices animals and birds, navigation stations the molecular sleeve bearing. The idea of attached to the body. Some of them may monitoring movements of glaciers, such systems came up around 1970. In even be placed in the body or eaten remote weather stations and pollution 1992 the US Senate got interested in the (edibles) for medical analysis and moni- monitoring and flood warning systems. idea because of the enormous impact the toring. These systems had several constraints technology will have on society – if it is such as price, size and lack of computa- realisable [8]. The combination of wearables and com- tional power. The new sensor technology puter networks is called wearable com- is making some of these technologies If efficient assembler systems – like the puting [4]. This is again part of what is ubiquitous, and enabling new applica- ones that exist in cells – can be made, the called nomadic computing. Some charac- tions to emerge. technology offers opportunities like mak- teristics of such systems are: ing food from garbage, producing cancer The strategic problems for the telecom- cell killing robots of the size of a blood • The core device is a network of com- munication industry are: cell and processors with molecules as puters; components. It is claimed that these tech- • Which of these applications will really • They may contain sensors of some nologies will be available sometime dur- take off? kind (GPS receiver, video camera, ing the next ten to thirty years. thermometer); • Do the sensor technologies require new network technologies? As we saw above, systems offering • Some or all of the computers may be access to information such as the Internet attached to a person, car, container or • How much traffic will they generate? create much telecommunication traffic. other mobile ‘platform’; • How will the traffic be distributed over The sensor technology requires applica- • The computers may offer interfaces the network? tions on fixed and nomadic platforms to which are context dependent (key- be connected together possibly creating • How should such traffic be priced in board, audio, screen) and which may even more traffic. The communication order for the applications to material- change as the context changes; requirement coming as a result of the ise? nanotechnology may be even much • They may require much communica- bigger. tion in web-like structures and on radio interfaces; 5 Nanotechnology The point of bringing in such far-fetched • Their applicability may be very sen- The idea of nanotechnology is to make technologies is that the commercialisa- sitive to the cost of communication. low cost manufacturing technologies on tion of the World Wide Web took us by the atomic level. Part of this technology surprise. We are just getting in touch is self-reproducing systems and assem- with the development of the sensors.
Traditional Internet Sensor Nano- technologies technologies technologies technology
Maturity Full Infancy About to Embryonic? be born?
Business Stagnated/ Big? Enormous? Astronomic?? potential declining?
Knowledge Well-known, Partly understood, Not understood Not understood and mastery fully mastered not mastered
Current Almost all Little Almost none None research focus
Required Little Much on applications, Much on applications, Some: at least know degree of little on technology much on inter- what it is and how research connectivity, much it develops. Must focus on technology have some degree of preparedness
Figure 4 Five strategic imperatives and existing and future products
52 Telektronikk 3/4.1998 Again the telecommunications industry References national Conference on Wearable may be unprepared. Nanotechnology is a Computing ICWC-98, Fairfax VA, natural evolution of these technologies 1 Audestad, J A. Telecommunication May 1998. and must therefore be taken seriously so and complexity. Telektronikk, 94, that we at least know what is going on. (3/4), 1998, 2–20. (This issue.) 6 Merkle, R C. Self replicating systems and molecular manufacturing. 2 Audestad, J A, Kjæreng, A, Mahieu, Journal of the British Interplanetary 6 Strategic Imperatives L. An object oriented model of the Society, 45, 1992, 407–413. telecommunications business. Telek- This leads us to Figure 4. Here we have tronikk, 94, (3/4), 1998, 54–61. (This 7 Merkle, R C. Design considerations shown five important aspects of the four issue.) for an assembler. Nanotechnology, 7, families of evolution we have discussed 1996, 210–215. above. 3 Baldwin, T, McVoy, D S, Steinfield, C. Convergence. Sage Publications, 8 Regis, E. Nano. Bantam Books, The most important point is that most of 1996. 1995. our focus today is on the traditional tech- nologies. In this area we – that is, the 4 Mann, S. ‘Smart’ clothing. MIT 9 Audestad, J A. How to survive in the whole telecommunications industry – Media Lab (http://www.wearcom. future. Telektronikk, 94, (3/4), 1998, direct almost all research efforts. Very org/smart_clothing/) 74–81. (This issue.) little is directed towards research on Internet applications. Still much of the 5 Mann, S. Definition of ‘Wearable research money in this area is on network Computer’. Presented at 1998 Inter- and protocol issues. Very little effort is put into sensor technology and the impact this technology may have on our systems.
Now the time has come to shift this focus, where most of the new research efforts should be toward Internet applica- tions and sensor technology. The tradi- tional technologies are more and more becoming cash cows, and as such we should avoid wasting our efforts on bringing the technology too far forward. In the current situation we cannot any- more afford to spend fifteen years on developing a technology like ISDN which is obsolete when it is introduced (in a funny way, the Internet demand for more bandwidth has resulted in an un- expected demand for ISDN). The reason why we should shift the attention in this way is that the new technologies will have enormous impact on what we are going to earn money from and how to conduct our business in the future.
Jan A. Audestad (56) is Senior Advisor for the Corporate Management of Telenor. He is also Adjunct Professor of telematics at the Norwegian University of Science and Technology. e-mail: [email protected]
Telektronikk 3/4.1998 53 An Object Oriented Model of the Telecommunications Business
JAN A AUDESTAD, ARVID KJÆRENG AND LAURENT MAHIEU
1 Introduction Subscription manager We saw in [1] that the organisation of telecommunications is becoming very Clearing house complex. The business may best be de- scribed by decomposing it into several roles. For completeness and easy refer- Content host Retailer ence the model is reproduced in Figure 1. Each box in the model is called a role which may correspond to a business area Contract manager of a bigger company or be a company proper. In the liberalised telecommunica- Content provider tions market more and more combina- tions are seen. This is also a result of the complex delivery of services beyond Platform operator System integrator simple person-to-person communication (Internet, transactions, outsourcing). Access provider We saw also in [1] that the telecommuni- cations business consists mainly of value Transport provider networks with both vertical and hori- zontal co-operation (or competition) as shown in Figure 2. In the figure several Infrastructure owner Figure 1 Business system companies are shown; some consist of many roles and some only of one role.
Subscription manager
Clearing house
Content host
Clearing house Contract manager
Subscription manager Content provider
Platform operator Platform operator
Access provider Subscription manager Retailer
Platform operator
Access provider
horizontal Transport provider vertical market Infrastructure owner Figure 2 Example company of co-operation/com- petition
54 Telektronikk 3/4.1998 These companies co-operate or compete fully defined by its speed, direction of ways are described in terms of different in delivering products to the market. We travel and location. For this problem it is objects. However, every person con- say that there exists a relationship be- irrelevant how big the motor is, how tained in the public register is described tween the roles if they co-operate or many passengers the car carries, the size in the same way; the same applies to compete. There are two types of relation- of the petrol tank, and what type of petrol every person contained in the personnel ships: vertical between different roles, injection system is used. Similarly, the database. Objects described by the same or horizontal between similar roles. The traffic lights are defined by which colour set of attributes are said to belong to the notion of vertical and horizontal rela- is shown and the cadence for shift be- same object type. All persons contained tionships is independent of whether the tween different colours. In this way in- in the public database are of the same roles are in the same or in different com- formation not relevant is suppressed in type. Each individual person is described panies. The difference between internal order to make the system easier to under- by a specific name, address, date of birth, and external relationships are only quali- stand. Moreover, the only thing that the social security number and marital status. tative, and the regulation ensures that if surroundings know about an object are We then say that each individual person one role is within the incumbent, the the attributes and their current value (or, corresponds to an instance of the object relationship is the same independently of what is equivalent, the state of the type. The concept of generalisation whether the co-operating role is within object). It is not shown how the object makes it possible to describe a large the same company or belongs to a com- changes state, that is, all such algorithms number of individual objects once only petitor. as well as other internal constructions are in term of the type from which it is hidden. instantiated (or brought into life if it is an The business structure describing all object that can be implemented in soft- operators within a given area is complex. If we take the windows system of a com- ware). Figure 2 is only a simple example com- puter, the attributes of a window object pared to what reality may look like. In may be all controls available for opera- The third important aspect of object ori- order to understand such complex struc- tions on the window (file management, entation is that, in models, objects with tures methods must be developed which editorial functions, size and location exactly the same attributes – that is, are hide complexity not relevant to the prob- management, acknowledgements, termi- of the same type – are equal and replace- lem but contain all information required nations, help functions and so on). The able. This is one of the consequences of for understanding the types of interaction operations are everything the user may the generalisation principle. In practise, and the content of it. The method pro- do with each attribute (use the mouse to this may not always be possible; a win- posed here is built on object orientation. move a window or request a given dows program may not be adaptable to action, or to enter a given set of charac- a given operating system of a computer ters). How these operations are per- because of constraints in the part of the 2 Principles of Object formed and which algorithms are used object which is invisible to the user. This Orientation are not visible to the user. This is one of has to do with construction practises and the most important advantages of object monopolisation of products, and has Object orientation is mainly a tool for orientation: information hiding. This nothing to do with the principles of software construction and is usually asso- mechanism enables us to distinguish be- object orientation. In the models we ciated with a programming language: tween what is publicly known about an make we will always assume that Simula, C++, Eiffel and Smalltalk. In object and what is private. The public exchangeability of objects is possible. definition of Open Distributed Process- view consists then of the attributes and ing (ODP) defined by the International how we can interact with them. This is Graphically we may illustrate an object Organization for Standardization (ISO), what the user of the object is allowed to as a rectangle as shown in Figure 3 [2]. object orientation is used also for defini- see, and contains everything required for The object is then defined by its type tion of systems in such a way that the using the object. In Windows this corre- name and its attributes. The operations specification is not directly transformable sponds to information contained in the we may do towards the attributes are to software. This is the way in which we user manual. The private view is all soft- generally of two types: we may read the will use the method here. ware, hardware and other elements value of the attribute, or we may change required to make the object perform its its value. For certain attributes it may not An object represents a concrete thing. tasks. In Windows this corresponds to the If object orientation is used to describe software construction of the system and road traffic, then cars, pedestrians, roads, is found in specification manuals in the pedestrian crossings, junctions and traffic company implementing this particular lights are all objects defined by their product. behaviour. The behaviour of the object is Type Person (reference) confined to the case where it is used, and The second important aspect of object consists of the definition of what it is orientation is generalisation. Every per- Attribute 1 Name John Smith composed of (attributes) and operations son is an object. For a given application Attribute 2 Address 1087 Route 66 (or methods) which can be used in order every person object contains the same Attribute 3 Birth date 12.12. 1958 to read or change one or more of these attributes, for example, name, address, Social sec. no. 123456789 attributes. The description contains only date of birth, social security number and Marital status Married such attributes which are relevant for the marital status in public registers; or problem at hand. In the example of road name, job type, skill set and performance a) Type definition b) Example of type c) Instance traffic, where the focus is to determine rating in a personnel database of a com- how queues are created, a car may be pany. A person described in both these Figure 3 Graphical notation of objects
Telektronikk 3/4.1998 55 be allowed to change the value of it, that shows an instance of the model, where Person Cardinality Person Relationship is, the write operation is not allowed. As John Smith has two children. There are Name Name an example the object type person of the other symbols for cardinalities: more Relationship Address Address public database is shown. In this object than one (see Figure 5), exactly n (indi- Birth date symbol Birth date Social sec. no. Social sec. no. type it may not be allowed to use the cated by the number n), zero or one write operation against birth date and (open circle). Marital status Is child of Marital status social security number. The instance is also shown. Note that the instance may A fourth important property of objects is be distinguished from the type by de- composition. This means that an object is a) Model picting it as a rectangle with rounded composed of several other objects called corners. components. The object itself is called a composite. Composition is also a rela- Objects do not stand alone in a model: tionship type. Since it is generic, it may they interact with other objects. This be convenient to use a special symbol for (reference) (reference) Is child of interaction is called a relationship. The it as shown in Figure 5. In this example John Smith Mary Smith relationship may also be complex and we indicate that a company may consist 1087 Route 66 I 1087 Route 66 contain attributes, that is, the relationship of several roles taken from the business s c 12.12. 1958 hil 10.10. 1988 d o is also an object. Like objects, relation- system. The company may then be 123456789 f 123455555 ships may be generalised into relation- instantiated as shown. Note that the Married Unmarried ship types. Relationships may be de- instance may consist of several roles of (reference) picted as shown in Figure 4. Here the same type, for example, subscription John Smith II two objects of the same type (person) managers delivering subscriptions to 1087 Route 66 are related by the relationship Is child different market segments. The 1+ cardi- 11.11. 1995 987654321 of. Note that we have introduced another nality against the role object type indi- Unmarried term called cardinality. The cardinality is cates that the company must contain at the number of objects a given object may least one role in order to be defined as a b) Instantiation be related to. In this case the cardinality company. of the person object on the left is one (every child must have only one father or Now we are prepared to look at the busi- Figure 4 Example of relationship mother); the cardinality of the object to ness system in more detail. the right (the child) is arbitrarily many, indicated by a filled circle, since a person 3 Example may have none, one or any other (reason- able) number of children. The figure also In Figure 6 we have shown a small part of the business system of Figure 2 in object oriented notation. The model has apparently not become simpler. When identifying the relationships this is usu- ally what happens. However, in order to Company (company) manage the model and find all relation- Name - ships, the model may be simplified. First Headquarter - we may delete the company objects and the composition relationship. This has Composition nothing to do with the business relation- symbol ship; this is organisation of the company. Cardinality Next we may split the model into inde- 1+ pendent parts where independence means that there is no direct relationship be- Role (subscription (subscription (access tween the parts of the model. Doing this, manager) manager) provider) Name we have to be very careful so that rela- Sales office - - - tionships are not overlooked. Products - - - Interfaces - - - Figure 7 shows an example of this Price list - - - decomposition. Here we have decom- posed the model into three submodels: a) Generic relationship sales, management and interconnect. By (network some training this way of handling the operator) model becomes simple, and we obtain - models which show exactly what we - want and from which we may draw - important conclusions. - Figure 5 Composition b) Instantiation
56 Telektronikk 3/4.1998 Company Company Sales cooperation
Subscription manager Subscription manager
Clearing house Sales cooperation Retailer Clearing house Sales Management
Subscription manager Platform operator
Management Sales Content provider Platform operator representative
Management
Platform operator
Interconnect Figure 6 Example of model
Content provider Subscription manager Sales Subscription manager Retailer cooperation
Sales cooperation
Sales Subscription manager a) Sales
Sales representative Clearing house Platform operator
Subscription manager
Management
b) Management
Platform operator Platform operator Platform operator
Interconnect Interconnect
c) Interconnect Figure 7 Decomposition
Telektronikk 3/4.1998 57 4 How to Identify Objects Interconnect and Relationships Telecommunication Telecommunication Mediator Mediator Here are some methods by which we may identify objects and relationships in a particular application of telecommuni- Single interface cations. Note that it is far too general to Customer make an overall model of telecommuni- cation businesses. The basic relations are found more easily for concrete examples. Market However, from such an analysis we will find that the same relationship objects are Figure 8 Symmetric mediation appearing several times. This means that it is not only for object types the model splits into a rather small number of dis- tinct objects. The same applies for the relationships. 4. The attributes of a role object define restriction of arbitrage, risk imbalance Some useful methods are as follows. The which interfaces other roles will have and business denial. Part of the analy- object types are those contained in the with that object. The activation of sis should be concerned with how business model in Figure 1. attributes will also depend on what products are complementary, 1. Decomposition of the company is the type of alliance or competition takes supplementary, or replace each other. starting point in the analysis, that is, to place with other roles. This means that This is certainly a dynamic process identify which roles it consists of and the interface may be different depend- and may not match a static regulation. which attributes are implemented in ing upon where and what a co-operat- the roles. This may uncover if there ing role is. 5 Value Creation are attributes lacking or being wrongly 5. This leads to a more general analysis defined, or if there are too many of alliances: which roles each partner Most of the telecommunication business attributes showing information about should have, roles shared between consists in mediating between different the company which should be hidden. partners and roles existing indepen- customers. Principally, there are three Note that the attributes are what we dently in each partner. distinctly different ways in which this will let the environment know about mediating activity takes place. the role. This may also uncover if the These are analyses of how a given company is not complete with regard telecommunication company performs its to role composition. Symmetric mediation business alone or in alliances. Concern- 2. Other roles which will not be imple- ing competitors similar considerations The market for telecommunications was mented in the company must then be apply: symmetric. This means that the commu- identified in order to see how we can nications were between entities of the 6. It is particularly important to identify include these roles in our business by same sort: users talking to each other the role composition of the competitor alliances. The alliances may be in- over the telephone, machines interacting and possibly how it may evolve. This ternal or external. This is one way of in order to complete transactions, and also involves analysis of which prod- completing the company with regard sending of documents such as e-mail and ucts such a role composition may pro- to roles and attributes since attributes telefax from one machine to another. duce. are also obtained by alliances. This gives a mediating structure as 7. The competitors may also form shown in Figure 8. There is only one 3. The company will make products alliances. interface towards the market and only where some products can be made one market. This type of mediation is entirely within the company while For the incumbent and for newcomers in similar to that of banking and insurance. other more complex products will the market it is important to understand require that many roles co-operate in the regulation of the market. The regula- This type of mediation exists both in the the production. Some of these products tion will be different for the two types of residential market and in the business may require co-operation far beyond actors. market. Interconnect is part of this medi- the role definition defined above, and ation service where interconnect ensures may in themselves require extended 8. In the business process not all inter- that all customers in the market can be models where for instance the content faces may be regulated nor which con- connected with each other independently provider may be decomposed. For ditions apply at the interface. On the of operator. each product it is, therefore, necessary other hand, interfaces which are in- to define which roles are involved and ternal in one company may be de- how these roles are divided among manded opened by the regulator for Simple asymmetric mediation internal and external relationships. forced co-operation. In this respect the This case is shown in Figure 9. In this Detailed analysis of how the roles problem is to identify what are equal case the mediation is between a ‘con- interact should be part of this exercise. conditions. The analysis should sumer’ market and providers of ‘goods’. include such things as possibility and
58 Telektronikk 3/4.1998 content provider. The interface toward the consumer is much the same as in Content providers the previous case.
The interface toward the content provider is different and may be so complex that it Content requires separate system integrators to interface build it. Payment for the services may be made via the telecommunications Interconnect operator, or be arranged in such a way Telecommunication Telecommunication that the consumer pays the content provider directly and the content provider Mediator Mediator pays the telecommunications operator. In this market we may see the first incen- tives to change the established payment Consumer and subscription structures. interface Consumer Indirect payment Direct payment Complex asymmetric mediation Market This case resembles the former case. The main difference is that the telecommuni- cations operator is one of several media- Figure 9 Simple asymmetric mediation tors interacting in order to deliver the service. The interface toward the content provider is then replaced by an interface toward other mediators. This is illu- strated in Figure 10.
Hidden structure The reason why this case is so important is that the telecommunications operator Content no longer has direct influence on the de- Content providers livery of content and the way in which Mediator telecommunication interacts with the product. Complex products like intelli- gent buildings, road traffic management Payments Business systems and co-operative work systems may be interface are examples of systems where telecom- uncorrelated munications is only one of many compo- Interconnect nents making up the product. Such sys- Telecommunication Telecommunication tems may also require several system Mediator Mediator integrators in order to establish the com- plex solution. It will also require co-oper- ation far beyond what is common today.
Consumer Note also that payment for the services interface Consumer may be made to some hidden content provider. The telecommunications opera- Market tor is paid back by the content mediator. This payment may be uncorrelated with the payment from the consumer. Figure 10 Complex asymmetric mediation The categories above are related to the interactions between markets and the providers of product elements beyond that of telecommunications. As we saw earlier, the telecommunications compa- Goods in this respect can be information astrologer. It is also the service which is nies are in addition split into roles, most of any type: video, music and text. What planned for video-on-demand and pay- of them performing mediating activities: makes it different from broadcasting is per-view. The mediation is illustrated in transport provider, access operator, plat- that the service is interactive. This is the Figure 9. The main point is that the mar- form operator, subscription manager, type of service that exists on the World ket for the telecommunication operator contract manager, clearinghouse and con- Wide Web. Premium rate services are splits into two markets, each with a dif- tent host. The characteristics of all of also simple examples where you may ferent interface: one market toward the them are that they deliver parts of the pay for advise from a lawyer or from an consumer and one market toward the overall product in parallel and at the
Telektronikk 3/4.1998 59 same time. The same do the other media- evolution of the common services and • Operation of mediating infrastructure tors and content providers in Figures 9 markets, and ensure that the quality of required for making the product avail- and 10. When talking of complex prod- the product is retained. In an alliance able to the customers. In the example, ucts like intelligent building, co-opera- consisting of energy producers, this activity may be shared among tive work and electronic trade, we are at telecommunications operators, installa- operators delivering communication the same time discussing a value creating tion companies and manufacturers of infrastructure for remote control and system more complex then any other equipment delivering remote control operators delivering support for trans- such system. of energy to the residential market, this action systems and information stor- activity may be assigned to a separate age. In such complex structures there seem to company in charge of binding the • Customer administration contains be two main activities: alliance together and ensuring that the activities like billing and customer right combination of skills is available • Development of the mediating infras- care. In the example, this may be han- for installation, marketing and opera- tructure; dled by several companies specialised tion of the system. in different activities such as billing, • Development of services in the mediat- • Development of core value and pro- maintenance, handling of complaints ing infrastructure. duction management is the core busi- and promotion. The activity should ness of developing a complete product. nevertheless appear as one single As was mentioned above, these activities The activity combines competencies activity. This may require an advanced may not be done by the value network in and technologies of all the actors tak- platform for co-operative work. the mediating infrastructure but by sepa- ing part in the production in order to rate system integrators. Once the infras- realise the final product. In the above In complex cases where the product is tructure has been established, there are example, this activity may be under- delivered by an alliance of several com- four generic activities creating the value taken by the producers of energy since panies each accommodating several in the structure: they will deliver the core element of roles, the basic activities described above • Common project organisation where the product. If the system also delivers may be distributed over several compa- the task is to divide the production a burglary alarm, the activity may be nies. Products may be distributed in among the different roles in the value shared among the producers of energy different ways, and the same companies creating network, and to support co- and the companies delivering the or parts of companies may enter several operation between the individual roles, safety component. alliances and even be present in compet- ing constellations. Examples of such complex constellations will be found in electronic trade, co-operative work, telemedicine, intelligent buildings, road control systems, and logistic systems. These are all areas hardly developed, where the development may have been hampered by poor understanding of the Customer underlying system of value creation. Core value management Note that the examples we are presenting on behalf of Content provider Subscription manager here are several magnitudes more com- (energy) establishes (billing) plex than simple mediating industries like banking, though division of this business into specialised roles may create on behalf of Subscription manager systems of comparable complexity. establishes (complaints) requires This division of responsibility may be components from found and implemented if an object for support Content provider oriented model is made of the organisa- of billing tion required for making the product. As (equipment) on behalf of distributes for we saw above, the object oriented model will contain all roles required in making the product and the relationships that System integrator must exist between the roles. A sketch of Platform operator (installation) how the activities may be mapped onto gets contract with (remote control) the object model is shown in Figure 11. uses This model only contains a rudimentary description of how a business system for Platform operator the intelligent building product may be (transactions) designed. The value production activities on behalf of System integrator (core management) Operation of may be allocated in different ways; this infrastructure is just one example. Note also that there Project may be activities which are not part of management the main value creation activities. Figure 11 Mapping of value cration mechanisms
60 Telektronikk 3/4.1998 6 Conclusions References
Above we have only given a glimpse into 1 Audestad, J A. Telecommunications a new way of looking at mediating indu- and complexity. Telektronikk, 94, stries or – what it is also referred to – (3/4), 1998, 2–20. (This issue.) value networks. The need for such tools is a result of several things: the increas- 2 Rumbaugh, J et al. Object-oriented ing complexity of telecommunications, modelling and design. Prentice-Hall, the new way telecommunication interacts 1991. with other products, and that some of these products can only be created by alliances combining several skills which have never been combined before.
The use of object orientation is one way of making the complexity comprehen- sible by only focusing on elements which are core elements of the subject matter, and by suppressing information which are not needed in the given context. Different ways of looking at the problem will require different models, and a com- plete picture of the system will consist of many models. This way of looking at systems is in fact why object orientation was first proposed in the early 1960s (the Simula language). In our opinion, object orientation is not just one way software Jan A. Audestad (56) is Senior Advisor for the Corporate can be constructed, but represents a much Management of Telenor. He is also Adjunct Professor of deeper approach to system design. This telematics at the Norwegian University of Science and approach is now gaining ground since Technology. ISO also uses the method for system de- e-mail: [email protected] scription, including whole enterprises, in their Open Distributed Processing speci- fication, and not only for design.
Arvid Kjæreng is Business Consultant at Telenor R&D, where he has been working with value creation processes in the ICT industry. His main focus has been to investigate business units’ positioning and business roles they can and wish to possess. He has also focused on value cre- ation alterations caused by the technical and content con- vergence. e-mail: [email protected]
Laurent Mahieu (28) is Advisor and Project Manager in the Corporate Strategy department of Telenor International AS. He holds a Masters degree in Economics from the University of Aix-en-Provence. In addition, he earned a M.Phil. in Systems Dynamics from the University of Bergen where he conducted research on the dynamics of strategic resources within the telecommunications in- dustry. e-mail: [email protected]
Telektronikk 3/4.1998 61 Real Options and Managerial Flexibility
STEIN-ERIK FLETEN, TROND JØRGENSEN AND STEIN W. WALLACE
The traditional net-present-value explain what real options analysis is and = a/2 + c = 5. We see that the entire (NPV) criterion in capital budgeting when it can be successfully applied. Sec- value of this project is due to the inherent has recently been heavily criticised tion 4 provides an overview and com- flexibility! (And we see that this is true because of its failure to take into parison of other tools that can be used in for all a < –b.) account the value of managerial flexi- investment situations, and Section 5 bility. In this paper we explain why provides some examples of investments The above example clearly demonstrates flexibility is important in real-life where managerial flexibility played a that the value of flexibility should be investments and discuss advantages role. Finally, in Section 6 we briefly dis- taken into account in capital budgeting and disadvantages of other approaches cuss how we can evaluate decisions decisions. While obvious in the above to capital budgeting, especially focus- made in the past. example, in more complex situations the ing on the real-options approach. value of flexibility is easily neglected. In fact, the classical net present value 2 Why managerial (NPV) criterion presented in most (if 1 Introduction flexibility is valuable not all) textbooks on capital budgeting is incorrect for projects where flexibility Many investment projects have flexi- By flexibility we refer to the ability to exists! Another way of saying this is that bility that is difficult to capture with make decisions at different stages in the NPV criterion is only correct in now- traditional discounted cash flow models. time. Generally, decisions depend on or-never investment situations (i.e. where The manager may have the option to what has happened at previous stages. all decisions must be made before the postpone the start of the project or to The absence of flexibility means that project starts). alter the project in some way either all decisions must be made initially. Of before or after the project has started. course, we will generally be better off A new emerging framework for capital This type of managerial flexibility be- if we can adapt the decisions during an budgeting, replacing the NPV paradigm, comes important for the total value of investment project rather than having to models flexibility in terms of options. the investment opportunity when there make all decisions initially. The differ- We discuss such real options in more is uncertainty in the environment of the ence in value between these two app- detail in the next section. project. The opportunity to defer a pro- roaches is called the value of flexibility ject is the most common example of a and it can be arbitrarily large as the real option, where the manager has the following example demonstrates: 3 Real options analysis right, but not the obligation, to initiate A financial call option is a right, but not a project at some future date. Suppose we have a risky investment an obligation, to buy an asset at a fixed project consisting of two stages such that price on a future date. If the holder of the Many investment projects can be viewed the second stage can be abandoned if the option chooses to buy the asset, often as real options. Such a view captures the first stage proves to be a failure. Thus, termed the underlying asset, we say that value of flexibility and is particularly in this project we have the flexibility of he exercises the option, paying the fixed suitable for situations where uncertainty deciding if the second stage will be exercise price to receive the underlying is important. In addition real options executed depending on the experience asset. This will happen only if the market analysis provides answers to when to use gained in the first stage. Let the profit price of the asset is higher than the exer- the flexibility and when to wait. (net cash flow) of executing the first cise price at the maturity (exercise) date. stage be c = –5 and let the profit asso- This paper is written for an investment ciated with the second stage be a = 20 if A real option is defined similarly, but is manager who makes capital budgeting the first stage was a success and b = –40 more general. The exercise decision is decisions for a firm facing an uncertain if it was a failure. If the second stage is not necessarily to buy an asset, but can environment. The manager is willing to not implemented, the profit associated be any decision that changes the nature formally analyse alternative investment with the second stage is zero. Of course, of the cash flows to the holder of the projects and to aid decision-making the optimal investment strategy in this investment opportunity. The irreversible through mathematical models. She or he simple case is to invest in the second investment cost that is committed at the is curious about which tools are available stage if the first stage is a success and initiation of the investment project often for such an analysis, strong and weak abandon if the first stage is a failure. If plays the role of the exercise price, and sides of these, and is particularly inte- there is a 50 % chance that the first stage the underlying asset is often the real rested in the real options view. is a success, the expected total profit is project once started. There may be no (a + 0)/2 + c = 10 – 5 = 5. So, the project particular date on which this investment As will be seen in this paper, the real gives a positive expected profit. opportunity matures, in which case the options view is not the only way to con- investment manager must find the best sider investment projects. Extensions of On the other hand, in the absence of flex- exercise date. discounted cash flow models and deci- ibility, we must decide initially whether sion analysis are related frameworks that both stages will be executed or not. We The most common types of real options can be employed, each having a different see that the expected total profit of exe- are: focus with different advantages and dis- cuting the project now becomes: advantages. (a + b)/2 + c = –10 – 5 = –15. Thus, the • The option to defer; project would not be started in the first • The option to abandon; The paper is organised as follows: In the place and we get a total profit of zero. next section we demonstrate why flexi- The value of flexibility is the difference • The staged investment option; bility is important, and in Section 3 we between the two cases: (a + 0)/2 + c – 0
62 Telektronikk 3/4.1998 • Scaling options; a discount rate of 12 % in their capital Project value (mill $) budgeting. • Switching options; 400 Value of now-or- • Growth options; The method used to solve the project never project value and optimal development strategy 300 Value with • Multiple interacting options. flexibility involves dynamic programming and Exercise at price solving a partial differential equation. We will next describe each of these in 200 $0.00036/minute The details are beyond the scope of this more detail. paper, however. 100 The option to defer the start Figure 3.1 shows the result of this valua- of a project tion. 0 0 0.0001 0.0003 0.0005 Even if a project has a positive net The dashed line shows the value of the Spot price ($/minute) present value (NPV), it may not be opti- project if the investment decision for Tel- mal to start the project immediately. For Figure 3.1 Investment strategy and value of cable Val is without the flexibility of deferring example, in a project consisting of con- concession for TelVal’s project the start of the project. The NPV is posi- structing and operating an oil platform, tive if the spot price is above $ 0.00018/ the value of the project depends heavily minute. This decision would only occur on the oil price. The future oil price is if the project had a positive NPV, which unknown, so the value of waiting can be happens above a spot price of $ 0.00018/ large. The reason for this lies in the minute. The solid line corresponds to the following asymmetry: If we defer the value of the project including the ability project and tomorrow’s oil price is suffi- project down! The reason for this is that to wait until the investment timing is ciently higher than today’s, then we will the future oil price (on which the value of optimal. The difference between these start the project. However, if the price the project depends) might increase to a two values is the value of flexibility. The drops, we do not need to start the project. level where the project again is profit- optimal timing of the initiation of the We have the option to defer and will wait able. The exact oil price when it is opti- project is when the spot price rises above until (if ever) the oil price is sufficiently mal to shut the project down, can be $ 0.00036/minute (which is twice what high. Generally, the value of this man- determined using a real options model. the classical NPV theory suggests!). agerial flexibility increases with the uncertainty. Thus, the more uncertain The staged investment option we are about the future, the more reason The option to abandon a we have to defer the project. project A common real option in projects with technical uncertainty (e.g. research and If an operating project becomes only development projects) is to divide the Example marginally profitable (e.g. the net present project into stages and make a go/no-go value is zero or slightly negative), the Consider a fictitious company, TelVal decision for the next stage of the project classical capital budgeting theory Inc., which has the opportunity to build based on the results obtained in the cur- suggests we close down (i.e. abandon) cable capacity to serve 300,000 long dis- rent stage. This introduces flexibility into the project. However, by real options tance telephone calls simultaneously be- the project and may enhance the value of theory we realise that abandoning a pro- tween two cities. This would cost $ 300 the project considerably. ject is a valuable option since it adds million, including the present value of all flexibility to the project. Since the NPV fixed costs. Suppose the company never criterion is invalid for projects with flexi- Scaling options has to give up the right to develop this bility, it follows that we must reconsider project. The management wants to know Note that all the real options mentioned the criteria for abandoning a project. The how much this project opportunity is so far come free of charge. In other option to abandon is similar to the option worth, and when to initiate it. words, these real options are inherent in to defer since shutting down a project can the projects and no additional investment be considered as an investment. So, even There is a spot market for use of such is necessary to have these options avail- if the net present value of future opera- cable capacity, with prices quoted in able (although there usually are costs tion is (slightly) negative (e.g. operating $/minute. Such a market exists, for associated with exercising the options). an oil platform where the costs of pump- example, at www.arbinet.com, and we However, in many cases it can be profit- ing up the remaining oil is very expen- will assume that these prices obey a cer- able to invest in additional real options. sive) it may not be optimal to close the tain stochastic process where the loga- For instance, the option to alter the oper- rithm of the spot prices follows a random ating scale of the project: For an addi- walk1. For simplicity, we assume that tional cost, the capacity of a new plant there is no variable cost associated with might be chosen at a late stage in the utilizing capacity and that interest rates construction process (of the plant) to are constant. The utilization rate is 85 %, catch the current trend of the demand. meaning that TelVal can sell Another important case where scaling 60 * 24 * 365 * 0.85 minutes per year options occur is in the general time-cost 1 The growth rate and volatility are on average at the spot rate. TelVal uses trade-off problem in project scheduling assumed to be 0.04 and 0.2, respec- where the project manager has the option tively.
Telektronikk 3/4.1998 63 to allocate more resources to any activity i.e. we cannot sum up the total value The DCF-based models assume the in the project before the activity starts. using many separate simpler models, but existence of several different cash-flow Thus, the amount of resources allocated need to have one single model integrat- scenarios, e.g. represented as a ‘decision to an activity will depend on what has ing the effect of all decisions made in the tree’. However, there are two major happened prior to the start of the activity project. Such complex models usually problems with DCF models. First, the (delays etc.) and will be chosen in a way require extensive numerical calculations, probability that a certain scenario will such that the project is completed on which in practice are only possible by occur must be specified explicitly (for all schedule. computers. scenarios). Estimating the probabilities is usually difficult. Second, DCF models Switching options depend on estimating risk-adjusted dis- 4 Alternative tools count rates. Under certain assumptions, A class of real options very similar to this can be done using the Capital Asset To maximise the value for shareholders, scaling options is switching options. For Pricing Model (CAPM), but it is very the investment projects with the highest example, a company might pay the addi- complicated in practical situations con- value should be selected. All decisions, tional cost of planning two mutually sisting of a large number of different both in selecting projects and in carrying exclusive versions of a product such scenarios. Simplifying the analysis by them out, should reflect this criterion. that only one of the plans can be imple- using only a single discount rate can lead Thus the investment and project man- mented. The choice of which plan to to large errors. agers should be making optimal deci- implement may depend on the current sions under uncertainty. In this situation demand. Examples of other switching DA models are similar to DCF models in it is natural to consider using stochastic options include the option of switching that both models can use ‘decision-trees’ programming techniques, a field which the technology of production or the for specifying cash-flow scenarios. Thus, focuses on optimisation and mathe- channels or targets of marketing the the problem of estimating the proba- matical modelling of problems having products/services that the real project bilities remains. The difference between uncertain parameters. supply. DCF models and DA models is the way risk is handled. DA models use the risk- Advantages of stochastic programming free rate of return as the discount rate. Growth options include modelling flexibility, simulta- This captures the time-value of money, neous valuation and optimisation of the With the classical capital budgeting but not the uncertainty. To model the project at hand and use of forecast scenar- theory there is often a contradiction be- effect of uncertainty DA models intro- ios. Disadvantages include the fact that tween investment decisions and company duces so-called utility functions and show no clue is offered on which discount rates strategy. A project might have a negative how a single decision-maker should act are to be used. A fixed rate is often used. NPV but still be recommended because to maximise the expected utility of the of its strategic importance. This paradox profit. This approach is based on certain Using a constant cost of capital, or dis- disappears when taking a real options axioms for rational behaviour of a single count rate, is also common in the dis- approach because it recognises that a decision-maker. However, a problem counted cash flow (DSF) method of project returns not only a cash flow gen- with this approach is that it is not market- investment analysis. In this method, one erated by the project itself, but also cre- based but rather based on the subjective finds the expected cash flows through ates new real options. The new options utility function of a particular decision- the life of the project, discounts them to are often referred to as growth options maker. In situations where a single de- present values, and adds them up. and allow the owner of the project to cision-maker cannot be identified or in execute follow-up projects. For example, situations where we want to maximise By varying the assumptions of which a project may give a company an option the market value of a company, we are information is available, different sto- to enter a new market which it otherwise outside the scope of DA models. chastic programming models can be could not. Such options can be very formulated. Capital budgeting models valuable and may outweigh the costs of CCA models (i.e. real option models) can be divided into three groups, depend- the original project. Growth options are overcome the problem of a single sub- ing on their underlying assumptions: ignored by the classical NPV theory and jective decision maker (in DA models) may explain why the market value of 1. Dynamic discounted cash flow (DCF) by assuming the existence of underlying some high technology companies (such models; tradable assets (e.g. commodities). Con- as Netscape inc.) has been (and probably tingent claims analysis (CCA) is based 2. Decision analysis (DA) models; still is) much higher than what a simple on analyzing a risk-free portfolio ob- NPV cash flow analysis suggests. Share- 3. Real option valuation models, also tained by purchasing one unit of the pro- holders appreciate the value of growth known as contingent claims analysis ject and selling a sufficient number of options even if NPV theory does not. (CCA) models. units of the underlying asset such that the variance (risk) of the portfolio is zero. Multiple interacting options All these models, if used correctly, can Based on this principle and using a take into account the value of flexibility. stochastic model describing how the In a general project, several types of real However, note that none of these models price of the underlying asset evolves over options are available. Most real option correspond to the traditional approach of time, CCA models can determine, among models where the solution is given by a estimating a single cash flow and then other things, the conditions when the pro- simple mathematical formula consider applying the net-present-value (NPV) ject should be started (as opposed to only a single type of options. However, criterion. We refer to the traditional mod- being postponed). The optimal condition the value of real options is not additive; els as NPV models. for when to start a project usually takes
64 Telektronikk 3/4.1998 the form of starting the project as soon as 5 Some real-life examples in case extra equipment is needed later the price of the underlying asset (e.g. for removing sand or water, or for pro- commodity price) reaches a certain level. Flexibility is generally important in all ducing at a higher rate. In order to employ contingent claims situations where we are making decisions • The cold war between the United analysis, the risk in the underlying pro- under uncertainty. We have already dis- States and the former Soviet Union ject has to have a market price. This is cussed different methods that can be used during the decades following the Sec- because the risk (or uncertainty) of the to deal with such flexibility (including ond World War is perhaps the most option depends on the risk in the under- real options valuation models). In this gigantic example where flexibility was lying project. In fact, the option derives section we present four, very different, essential. Both sides purchased the its value only because there is uncer- real-life cases demonstrating the exis- option to annihilate the other by install- tainty in the underlying project. The risk2 tence of flexibility: 1) in copper mining; ing intercontinental ballistic missiles is priced, for example, if the project is to 2) in the construction of a power plant; (ICBMs) armed with nuclear warheads. produce a storable commodity sold in a 3) in the design of offshore oil platforms; From a military point of view, having perfect market, and the only important and 4) in the cold war! this option had a large positive value source of project uncertainty is the future • Canadian copper mines have been because it would prevent the other side price of the commodity. If the product or extensively studied using real options from attacking. In this case, formal real service is not storable, there may still theory [4]. Copper mining is a risky options theory is not applicable since it exist a futures or options market having activity due to the very volatile copper is not a market based problem. How- the price of this product as the underlying price. However, if the copper price ever, real options theory gives us a use- reference, in which case the price of risk gets too low, the management has the ful terminology for describing the situ- is reflected in the prices of futures or flexibility to temporary shut down the ation. A more suitable formal approach options. A practical problem with CCA mine. Later, when the copper price is in this case is applying game theory models is obtaining a realistic stochastic again sufficiently high, the mine can (i.e. an extension of decision analysis model for the price dynamics and the reopen. This option of temporary that deals with problems where the out- price of risk. This can be done by ana- closure is neglected by traditional come is affected by several decision- lyzing historical data via statistical NPV models. makers as well as uncertainty). methods. Choosing an arbitrary sto- chastic model can lead to arbitrary • The Svartisen Power Plant was de- results. Thus, back-of-envelope calcula- veloped by Statkraft from 1987 to 6 Evaluating decisions tions using simple formulas obtained by 1997. It is based on hydro-electric using a particular stochastic model, are power and has a generation capacity of made in the past dangerous3. Another practical problem 350 MW, which means that it must run Both in the power-plant example and the with CCA models is obtaining the rate of for about 7000 hours on average each cold-war example the options purchased foregone earnings. For financial options year to avoid reservoir spill. The ori- were not exercised (to date). Looking this corresponds to the dividends paid. ginal plans were to have twice this back in time and evaluating decisions capacity, 700 MW, to go with the large made, it may now be obvious that signifi- In all the three models – DCF, DA and reservoir capacity. However, Statkraft cant costs of purchasing such and similar CCA – stochastic programming tech- used the flexibility they had available options could have been saved if we had niques give the tools needed to obtain to adjust the generation capacity after known that the options would never have the optimal decisions under uncertainty. initiating the project and after seeing been exercised anyway. By making this Stochastic dynamic programming is a the changes in the profitability (but argument, it may seem tempting to con- good example of such a tool. In addition, before the end of the development). In clude that the options should not have advanced statistical methods are used for other words, they exercised a scaling been purchased. However, making such a describing the dynamics of prices. A lot option. At the same time, they decided conclusion would not be rational! At the of effort has been made to develop ana- to buy the flexibility to reverse this time the decisions were made, the deci- lytical solutions to capital budgeting decision by reserving space in the sion-makers did not know if they would problems for important special cases. power station house and building exercise the options or not, due to the However, a numerical approach (using a tunnels for installing a second 350 MW presence of uncertainty. In retrospect, the computer) is necessary for more general generator at a later date. In real options only way we can find out if an optimal problems. terminology: They purchased a growth decision was made is by analysing the option. Although real options theory situation using only the information was not used, the decisions were cer- available to the decision-maker at that tainly subject to formal analyses at point in time. The scenario that actually Statkraft. did occur, was only one of several pos- • Design of offshore oil platforms must sible at the time the decision was made. be done under the uncertainty of the It was, of course, beyond the decision- nature of the reservoir. There can be maker’s control to choose the scenario! 2 Or more precisely, the part of the risk more sand or water in the reservoir If the information available for the deci- that is impossible to diversify away than expected, the size of the field is sion-maker is not available in a later through other investments. unknown, and future oil prices are evaluation of the decision, we cannot uncertain. Thus it may be profitable, know if the decision was optimal or not, 3 This conclusion is supported by analy- although very expensive, to reserve even if we know what happened later! sis regarding the profitability of Cana- some blank space on such a platform Thus, learning from history is not trivial. dian copper mines [4].
Telektronikk 3/4.1998 65 7 Conclusion References 4 Slade, M E. Managing projects flex- ibly : an application of real-options In an uncertain environment flexibility in 1 Christiansen, D S, Wallace, S W. theory. Vancouver, Univ. of British investment projects becomes more valu- Option theory and modeling under Columbia, Dep. of Economics, 1998. able, and for managers to capture these uncertainty. Annals of Operations (Discussion paper No. 98-02.) values and make better decisions it is Research, 82, 1998. (In preparation.) necessary to employ models that incor- 5 Trigeorgis, L (ed.). Real options : porate both uncertainty and sequential 2 Dixit, A K, Pindyck, R S. Investment managerial flexibility and strategy in decision-making. Because investments under uncertainty. Princeton, NJ, resource allocation. Cambridge, MA, have a strategic importance for firms, Princeton University Press, 1994. The MIT Press, 1996. those firms that are able to utilise and acquire flexibility wisely will have a 3 Kall, P, Wallace, S W. Stochastic better chances of surviving in the long programming. Chichester, John run. Wiley, 1994.
Stein-Erik Fleten (27) is a graduate engineer from NTNU 1995. He is a resarch fellow at the Department of Indus- trial Economics and Technology Management at NTNU and is working on a Ph.D. thesis with the working title “Theory of portfolio management in the electricity in- dustry”. e-mail: [email protected]
Trond Jørgensen (28) is a doctoral student in Operations Research at NTNU. He holds a Master’s degree from the University of Oslo, 1994. He has worked 3 years at Nor- wegian Defense Research Establishment in the Satellite Remote Sensing Group. He also has experience in un- supervised classification in multispectral satellite images. Other jobs include computer programming and teacher in computer programming at high-school level. His main interest is stochastic dynamic decision problems arising in complex systems. e-mail: [email protected]
Stein W. Wallace (42) has been Prof. of Operations Re- search at NTNU since 1990 after spending 7 years as Senior Scientist at Chr. Michelsen Institute, Bergen, and 3 years at Haugesund Maritime College. Throughout, his focus has been on theoretical and applied aspects of decision making under uncertainty, with applications in fisheries management, petroleum field development, portfolio management and telecommunications. e-mail: [email protected]
66 Telektronikk 3/4.1998 New Aspects of Service Provision and Technology Strategies in Telecommunications
SHANE DYE, ASGEIR TOMASGARD AND STEIN W. WALLACE
Because of deregulation and new video conferencing, it is highly unlikely 2 Distributed processing technology the telecommunications that the future will hold telecommunica- markets are changing. As the com- tion networks without transportation bot- in telecommunications puter industry and the telecommuni- tlenecks. However, many new services Aspects of distribution are crucial to the cations industry become more similar, and some transportation technologies modelling in later sections. Here, we give services based on processing of infor- require more and more processing a brief description of the distributed mation become more important than resources, thereby creating a new bottle- framework used for the discussion and before. This means that there is a need neck. Add to this the computing indu- the models of later sections. for new technology strategies and new stry’s influence upon the telecommunica- decision support tools at both the oper- tion market, and you get both additional ational and the strategic level. In this computational power and competition. 2.1 A distributed network paper we describe a new approach for Together, this makes it difficult to predict model decision support at the operational which of the bottlenecks will prove to be level of service provision when services most influential in the future. The relationship between the transporta- are processing based. We particularly tion network, the computing nodes with treat the case where demand for ser- There is much ongoing research investi- processing resources and the applications vices is uncertain, discussing how the gating ways to increase transportation used to build services can be represented flexibility of new distributed process- capacities and make efficient use of the by Figure 1. ing environments can be used to deal transportation resources available. How- with this. Strategic effects of such an ever, while there is a push to provide The model consists of three planes. In the approach are briefly discussed. An increased processing resources there is upper plane we have the set of interacting example model is given to illustrate little research into the efficient use of applications which form an interconnect- resource allocation in a distributed available processing resources. This is ed structure. Links here are representa- processing environment. the issue we address. In particular, we tive of the interactions between the appli- review the situation where computational cations in order to provide services. In resources provide the only bottleneck. 1 Introduction When considering processing resources New possibilities for providing services there is little telecommunications litera- in distributed telecommunications net- ture. In [12] the distribution of the work- works have appeared as a consequence load in a network of computers is dis- of technological developments such as: cussed with whole jobs being processed digital technology, modern packet on a single machine. In [18, 19] modell- switched high speed networks, architec- ing of distributed processing in service tures like ATM [4, 16] and standardiza- provision is examined. Related to the Applications tion of software and equipment. See [13] models given there, an algorithm to find for a discussion on the technological a feasible service configuration for the aspects of distributed processing in net- case where demand is known and all works. While [13] does not focus on demand for services must be met is given telecommunications, the in [8]. Approximation algorithms for the Telecommunications Information Net- case when demand is known and profit is working Architecture Consortium to be maximized are given in [5]. The (TINA-C) [2, 17] is an example of an ini- only known approximations for the case tiative defining a future distributed when demand is uncertain is when ser- telecommunications standard. vices are installed only at one node [6]. As we will see later in Subsection 4.3 The new technological aspects have not this single node case has specific and only introduced new possibilities in important interpretations also in a dis- telecommunications but, along with tributed processing setting. deregulation and free competition, they have contributed to a change in the envi- This paper is based on the results in [5, 6, Computing nodes ronment where the telecommunications 18, 19]. Section 2 defines the technologi- industry finds itself. Competition has cal framework we operate within. Then become harder; new players enter the we look at service provision in Section 3. scene and the roles of old players change. The main focus is on an operational approach for service provision trying Traditionally, switching and bandwidth to utilize the flexibility of a distributed are considered important bottlenecks. processing environment. The strategic Currently, transportation capacities are effects of such an approach are then dis- being increased through the use of fibre cussed. In Section 4 we examine a model optics, advanced protocols and data com- to illustrate the service provision app- Networks pression technology. With the parallel roach and give some results on the hard- emergence of new services such as video ness of such models. Figure 1 The relationship between the transportation on demand, pay per view television and network, the computing nodes and service applications
Telektronikk 3/4.1998 67 the bottom plane we have a set of inter- ing the customer’s service or other vices can be distributed, with their sub- connected networks describing the physi- interacting applications. service instances spread over several cal connectivity between the applica- nodes. Figure 2 illustrates this for service tions. The computing nodes, where the Such properties will be implied by many B. Also several instances of the same applications reside, can be seen as the distribution frameworks. In particular, subservice may exist at different nodes, ‘glue’ which binds the distributed appli- they are implied by the transparencies of as for subservice ‘sound’ in the illu- cation architecture and the network archi- TINA-C [1, 3, 15]. strated example. The arcs in the lower tecture together. The (all-to-all) links in half of the figure represent the different this plane show which computers have A further assumption about the distri- instances. A subservice instance can be access to each other through the under- bution that we make use of in various used by several services and a service lying networks. models is usually consists of several subservices. • The processing of an application may Note that the mappings between the three be provided externally without affect- planes may be many-to-many mappings, 3 Service provision, ing the customer’s service or other i.e. one computing node may accommo- interacting applications. network design and date several applications and several identical copies of an application may distributed processing Applications which are provided ‘exter- reside at different nodes. Similarly, one nally’ might be provided by a different Firstly, we discuss more specifically the computing node may be connected to part of the service provider’s network or task of allocating network resources to several networks, and one network may by a different service provider. services in order to meet demand. We accommodate several nodes. Further- start with describing two types of re- more, these mappings can be dynamic. sources – node resources and transporta- Applications may be installed or re- 2.3 Distribution of services tion resources – before we examine a moved from computing nodes. Also, resource allocation approach utilizing the systems are emerging where even the We define a subservice as a collection of network’s distributed processing capa- mapping between nodes and networks applications that will always be located bilities. Thereafter, we study how can be dynamic; for example, in the form together to perform their assigned task. resource allocation at the operational of computing nodes residing in mobile In the rest of this paper we will consider level influences network design. terminals or low orbit satellites. the subservices as the smallest indivisible entities of computer code that are the It is worth noting that this distributed net- building blocks for the construction of 3.1 Service provision and work model provides a useful model for services. limited resources the Internet and World Wide Web appli- cations. The upper plane corresponds to Therefore, all services are built from sub- A service is built up by interacting the Web, the middle plane represents the services, as shown in Figure 2. Here ser- subservices. The subservices and their servers, and the bottom plane is the inter- vice A is built from subservices ‘sound’ interaction require both transportation national structure of telephone and data and ‘video’. Service B is built from sub- and computational resources. Typical networks. services ‘sound’, ‘data’ and ‘video’. The transportation resources are bandwidth, distribution transparencies mean that ser- switches, servers, transmission options 2.2 Distribution implications
The exact technicalities of the distributed environment are not important for the modelling discussion. Instead, we base our models on a framework implied by the distribution model. The important aspects are as follows. service A service B services composed • Applications which together constitute of subservices a service can be placed on different computing nodes without affecting the interaction between them. • Applications may be dynamically re- located, even whilst providing process- ing and interaction with other applica- video sound data subservices tions. • Several copies of an application may be present in the network. Figure 2 Services • The processing of a single customer computing nodes are composed of (or customer group) may be dyna- plane subservices which mically moved from one copy of an are distributed in application to another without affect- the network
68 Telektronikk 3/4.1998 and routers. Computational resources When computing capacity is not the only the choice of how much to invest in each are provided at the network’s computing limiting resource, these assumptions pro- of these resources. Then, remember that nodes; examples include processing vide a conceptual framework for decom- within a region like the ones defined in capacity, memory and input/output posing a global network into smaller sub- Assumption A2 of Section 3.1 the cus- capacity. networks for which the following models tomer is indifferent to where demand is are valid. met. This allows for ‘trading’ of process- The customer’s requirements from a ser- ing capacity between different locations vice are covered in the concept Quality of It is also interesting to note that the mod- of the region: processing capacity in- Service (QoS) [10]. QoS for applications els may remain useful even without stalled at one location can be used to or subservices can be looked at as a set of assuming that computing resources are meet the demand from several other loca- user perceivable attributes defining the the bottleneck. This is because the net- tions. This is useful when different parts desired behaviour of the subservice when work provider is not necessarily the ser- of the region experience peaks at differ- it comes to, for example, delay, reli- vice provider. A service provider may be ent times of day, like for example a city ability, image quality, image size, secu- free to rent computing capacity from the in business hours and its suburbs after rity etc. At a lower level one also defines network provider. In the same way that work. In addition trading gains benefit QoS parameters for the networks: band- the location of service processing may be from the economies of scale and utilizing width, delay, blocking probabilities, loss hidden from the customer, the transporta- resources in low investment cost areas. If rates, etc. In light of the properties of tion structure may be hidden from the demand is uncertain and not completely Section 2.2 we assume that the QoS of a service provider. Then the service pro- correlated in the different parts of a subservice is not affected by the specific vider’s limiting resource is solely the region, trading of capacity is one effi- computing node on which it is installed. purchased computing capacity. cient way of dealing with the uncertainty. Capacity trading of course increases the The properties from Section 2.2 indicate The main conclusion we can draw from need for transportation resources, as that the location of the subservice in- the above discussion is that given the information must be moved before it is stances used to meet requests is not assumptions, customer demand for pro- processed. Even if trading of capacity in important from the point of view of the cessing capacity can be met at any com- a distributed processing environment interacting subservices. Additionally, puting node independent of the location means more flexibility and is one of the customers have no preferences when it of demand. most important tools to deal with uncer- comes to the service instances they use, tainty both at the operational and the except through QoS. The distribution 3.2 Network design and critical strategic level, the extra overhead in therefore gives rise to a great deal of resources in a region transportation must be considered. flexibility when it comes to allocating resources in the network, and are crucial Additionally, network designs which The flexibility of the above approach at to the problem described in this section. tend to use the transportation resources the operational level also has strategic most efficiently often make processing a effects. The choice as to which resources We make the fundamental assumption bottleneck, and vice versa. This is illus- limit a telecommunication system is not that we have a distributed network in trated in the following extreme examples. arbitrary. The network’s design is vital to which the properties from Section 2.2 are providing the balance between network valid. If processing capacity is the only Since the network is distributed by resources. Here, we provide a brief dis- limiting resource, the following assump- nature, control information and signalling cussion of network design. We focus on tion makes sense: are reduced as a consequence of a dis- the degree of distribution of the process- tributed architecture. This reduces the ing capacity at the strategic level and its A1: In regions of the network, a ser- transportation load. An extreme possi- effect on flexibility at the operational vice’s demand for processing bility, therefore, is to use only local pro- level. capacity can be served at any cessing with many small computing computing node in the region, nodes. The effect is little transportation, The discussion here is to highlight the independent of the demand’s and but low utilization of processing capacity trade-off between transportation and the node’s locations and without as each subservice will be installed in processing from the point-of-view of net- affecting the service’s QoS. many locations. Typically there is an work design. Given this perspective, we overhead in terms of resource use con- do not include existing network resources This in turn implies the following under- nected to running a subservice, for ex- in the models. In practice, the current lying assumption: ample a fixed use of capacity indepen- infrastructure must be considered and it dent of the number of customers. This brings to the models additional restric- A2: Within regions of the global net- overhead will now be induced every- tions relating to such things as mixing work, the transportation networks where. At the other end of the spectrum, new and old technology, keeping ser- are designed to keep the time delay using one enormous, centralized com- vices available while improvements are acceptable on transportation of puting node only requires one copy of being made and legal aspects restricting generated information and to each subservice. However, signalling or requiring change. ensure the correct QoS for ser- increases and all services must send vices, independent of the alloca- information through the centralized node, Finding an ‘optimal’ network design is tion of requests for subservices both of which create a bottleneck in not straightforward. The trade-off be- to nodes. transportation resources. In addition, the tween transportation and processing centralization raises reliability issues. resources is complex. Primarily, there is
Telektronikk 3/4.1998 69 We try to describe trading of processing possible through the use of distributed node, even when it does not satisfy a sin- capacity and the trade-off between pro- processing. gle request. The second type of capacity cessing and transportation capacity using use comes from accessing the subservice. an hierarchical approach to network This requirement increases in direct pro- design by considering two models. The 4 A service provision portion with the subservice’s demand ‘local’ model determines investment in model met at the node. The combined capacity network infrastructure (telephone lines, use of a subservice is illustrated in Figure switches and computing nodes) within a We will now present a modelling app- 3. (small) region of the network. Typically, roach to service provision of processing this would be a region as defined in A2. based services as described in Section 4.2 A deterministic model for This requires the infrastructure to allow 3.1. Afterwards we discuss uncertainty, several computing nodes the chosen demand to be met with a how to deal with it, and some results on given level of QoS. The increased trans- computations. This section is based on We first describe a static model assuming portation infrastructure cost from trading [5, 6, 19]. deterministic demand. The main reason processing capacity within the region is for considering such a simplistic model is weighted against savings in installing 4.1 Assumptions that it gives us insight into the underlying computer capacity. allocation problem by being easier to To provide a clear and consistent model analyse. These insights will provide use- The ‘global’ model determines how to we make the following assumptions. The ful guidance for more realistic but more invest in network resources and allocate number of requests for a service de- complicated models where the demand demand among these various regions. scribes the demand from the customer fluctuates over time and is not known in We also allow trading of processing locations. Demand for a service induces advance. resources between regions, again at the a corresponding demand for the subser- expense of transportation resources. vices used by the service. The demand The model we present is a linear mixed Clearly, in this model we implicitly induced for a subservice differs from integer programming problem (MIP) assume transportation between regions service to service. In the model we only [14]. The aim of the model is to best allo- to have a limited effect on QoS. This is implicitly consider the service demands cate the subservices to the computing valid as the transportation links are through their combined demand for sub- nodes. Subservice demands which are planned for meeting the extra demand services. We allow subservice demands not met by the computing nodes are met for transportation capacity that is induced to be met either internally or externally. externally and we define the best alloca- by trading. These models are detailed in A demand met internally is processed by tion as having the least cost for demand Tomasgard et al [18, 19]. a subservice at one of the computing met externally. There are m computing nodes. A demand met externally does not nodes and n subservices. The zero-one For the remainder of this paper we use any node resources. Instead, there is variable z is 1 if subservice j is installed implicitly assume that the investment ij a single, linear cost associated with the on computing node i and 0 if not. The problem is well solved and that we are externally met demand. In practice, this variable x denotes the amount of free to utilize the computational ij demand will either be met in another demand for subservice j met at node i, resources in the network as efficiently as region of our network or we will pay while tj is the demand for subservice j another service provider to meet this met externally. Let the demand for pro- demand for us. In the latter case a spot- cessing capacity for subservice j be dj. market exists for trading subservices. The cost of processing subservice j ex- One can imagine several types of compe- ternally is qj. The fixed resource use by tition. We will here assume that none of subservice j is denoted by rj. The total the service providers are able to influ- capacity of node i is si. M is an arbitrarily ence the price by their decision, and large number which provides an upper regard the market price as fixed. If a bound for all xij. We formulate the model resource more complicated price structure is as follows: assumed, the models given here will n be a crude approximation. min qjtj j=1 When a subservice is installed on a node, n n the only limit on the number of requests s.t. rj zij + xij ≤ si,i=1,...,m, that can be served by it arises from the j=1 j=1 node’s processing resources. In respect to m this, there is one constraining processing tj + xij = dj,j=1, ..., n, resource, referred to as processing capac- i=1 ity. We distinguish between two types of fixed processing capacity used by a subservice. Mzij − xij ≥ 0,i=1, ..., m, j =1, ..., n, amount The first is a fixed amount (called the zij ∈{0, 1} ,xij ≥ 0,i=1, ..., m, j =1, ..., n demand met fixed resource requirement) induced by having the software running and ready to tj ≥ 0¯v =1, ..., n (1) meet requests. A subservice uses this Figure 3 A subservice’s use of processing capacity capacity whenever it is installed on the with increasing demand met
70 Telektronikk 3/4.1998 The objective function calculates the changing the resource allocation. This the first stage. This decision process is total cost of meeting demand externally. ability to change is even more important illustrated in Figure 4. In the figure, at The first set of constraints ensures that when we try to consider uncertainty of the point where demand is revealed, only the computing node capacities are ad- demand in our models. In the treatment one of the two scenarios S1 and S2 will hered to for each node, while the second of uncertainty here and in the rest of the be realized. We cannot know beforehand set of constraints requires that demand is paper, we use terminology from stochas- which path will occur. either met externally or internally at tic programming [11]. In a stochastic some computing node. The third set of programming setting we assume that The current value of a given first stage constraints forces nodes which meet the uncertainty in the demand for services decision is found by evaluating the cost demand for a subservice to have that sub- can be modelled as random parameters of meeting demand in the second stage service installed. with known probability distributions. for each modelled scenario in the best possible way, given the fixed first stage 4.3 Different interpretations of At any time the state of the system of decision. For each first stage decision, the model computers running services can be de- we see a probability distribution of sec- scribed by the current demand for pro- ond stage costs. The optimal first stage cessing capacity, a probability distribu- decision has the least expected second In this model we implicitly assume that tion for future demand and the current stage cost. For a further introduction to the service providers have full knowl- whereabouts of subservices. In this paper modelling of dynamic decision processes edge of the sizes of all the computing we focus on the case of discrete distri- using scenarios see for example [20]. nodes and the ability to utilize them as butions; each realization of demand is they see fit. This may be the case, for called a scenario [11] and can be inter- example, if the service provider and net- 4.5 Heuristics and com- preted as a demand vector with an work provider are one and the same. putational complexity assigned probability to occur. When the service provider is not also the of the models network provider, the situation differs When a decision is made to install a sub- depending on the level of control allowed In the previous section we advocated the service at a node, the subservice cannot to the service provider. The service incorporation of uncertainty in the model usually be used to meet the demand provider may just hire a certain guaran- without indicating how one would solve immediately. There is a set-up time in teed quantity of processing capacity from the resulting model. In fact, adding which the subservice must be collected the network provider. The service uncertainty to a model generally makes it from a database, loaded into the memory provider can decide what subservices to more difficult to solve. The single node and initialized, before it can be used to provide and levels of demand to meet, deterministic problem (m = 1) can be serve requests. There can also be a delay however, the network provider is free to solved efficiently by dynamic pro- between the time the service provider replicate and migrate the subservice gramming [5]. The algorithm’s running decides to remove a subservice and the instances running on his infrastructure. time is pseudo-polynomial, that is the time its fixed resource use is released. The service provider sees the infrastruc- number of steps is bounded by a poly- Under specific assumptions [19] con- ture only as a single node. The above nomial in the number of subservices and cerning set-up times, shut-down times model is still valid, but now m = 1. Sin- the node’s size. However, the multiple and the uncertainty structure, the de- gle node models are discussed in [5, 6, node deterministic problem is hard to cision process can be modelled as a two- 18, 19]. solve to optimality. Typically, it requires stage process. In the first stage we decide a strictly exponential method such as which subservices to install on which However, the network provider must still branch and bound. The stochastic mul- computing nodes, having only probabi- solve a multi-node problem. By hiding tiple node problem is at least as difficult listic information about demand. During the complete infrastructure from the ser- as the deterministic. Decomposition the subservices’ installation time un- vice provider the network provider is methods are used in [9] to accelerate the certainty is resolved. At the second stage completely free to best utilize his re- solution process. For the stochastic single we must meet the realized demand using sources, possibly combining the needs of node case no pseudo-polynomial method the subservice configuration chosen in many service providers. The cost to the is known, but decomposition based on network provider of such a scheme is that efficient algorithms for subproblems may the implicit costs of subservice replica- even here be a promising approach to tion are not directly borne by the service solve the problem efficiently [7]. If we provider. His gain is that he is free to xS1 consider the number of scenarios as fixed reuse subservices for several service and not as a problem input, even this ∼ providers. z δ problem can be solved in pseudo-poly- nomial time. 4.4 Modelling uncertainty xS2 It is clear that currently there is no hope of obtaining an algorithm that solves the Understanding the effects of uncertainty time service provision problem to optimality on a system is an important step to fully set up demand meet in a real-time decision system. Indeed, understanding the system. subservices revealed demand theory suggests that there will always be a limit to the size of problem that can be One of the main advantages offered by Figure 4 An example of the two-stage solved in reasonable time. Consider that the distribution transparencies of Section decision process. Here there are two currently a deterministic single node 2.2 is the possibility of dynamically possible scenarios, S1 and S2
Telektronikk 3/4.1998 71 model with around 250 subservices will then chooses whether to install all sub- change in focus from transportation to take up to a few seconds on a Sun Ultra 2 services up to the critical one or simply processing for many new services, such machine running at 200 MHz. Similar install the critical one alone. This gives models should include distributed pro- sized deterministic multiple node prob- a solution that is at least half the optimal cessing oriented aspects of a telecommu- lems or single node problems with solution. The maximum running time is nications network. One reason for this is stochastic demand have less predictable proportional to the number of subser- that the computer industry and the solution times because of the methods vices. For the multiple node problem the telecommunications industry become used, but they typically vary between nodes are considered in any order. The more and more similar. seconds and several hours. critical subservice for the current node is left out and the next node is filled. The Given the assumption that processing A real-time system would need to use heuristic chooses whether to install this capacity is the limiting resource in our heuristics or rules of thumb as decision solution or one of the critical subservices network, we conclude that, in regions support. These provide fast but non-opti- singly. The greedy solution is at least of the network, allocation of computer mal solutions. The trade-off is between 1/(m + 1) times the optimal solution. resources to services at an operational quality and speed of solution. In practice, The maximum running time is propor- level may be made independent of the even so called ‘optimal solutions’ are not tional to the sum of the number of sub- location of the demand. We suggest how always the best, as models never pre- services and the number of computing the increased flexibility coming from dis- cisely portray reality. So the trade-off nodes. tributed processing and trading of capac- between speed and quality is not always ity can be used to dynamically allocate possible to accurately estimate. For the stochastic single node problem, a resources in a distributed network. similar heuristic is based on solving the One method for evaluating the effective- linear programming relaxation of the Increased flexibility clearly influences ness of a heuristic is to consider their problem. The solution time is still poly- the decisions to be made for investments worst-case performance ratio, a measure nomial. Here we choose to install all sub- in infrastructure, both when it comes to on how badly they can perform. The services for which zj = 1 or any of the processing and transportation capacities. heuristic solution is measured against the subservices where 0 < zj < 1 at the opti- Strategic decision models for distributed optimal solution. To compare heuristics, mal solution of the LP relaxation. This telecommunication networks cannot be a bound on the running time is also used. leads to the heuristic being at least studied isolated from models concerning This allows users one way to evaluate the 1/(k + 1) times the optimal solution dynamic resource allocation and the trade-off between speed and quality. where k is the number of scenarios. More assumptions they are based on. A shift in Good heuristics or approximation advanced heuristics with a constant per- focus at the operational level from trans- methods give solutions that are no more formance ratio exist even for this prob- portation of data to processing of infor- than a known constant multiple of the lem, where the heuristic solution can still mation is also bound to present a need optimal solution, with running times be calculated in polynomial time. for modelling new aspects when it comes bounded by a lower order polynomial. to investments. For the deterministic single node service We now discuss some previously studied provision problem it is possible to app- heuristics for the service provision prob- roximate in polynomial time the solution References lem from [5, 6]. For this discussion we value within 1 – ε of the optimal solution 1 Audestad, J A. Lecture notes : consider a slightly changed problem. value, utilizing dynamic programming. Course 45356 Communication in Instead of minimizing the cost of ex- Here ε can be chosen to be as small as distributed systems. Trondheim, ternal servicing, the profit from internal desirable, but the solution time of the UNIT/NTH, Department of Com- servicing is maximized. Here the profit approximation algorithm increases in 1/ε. puter Systems and Telematics, for meeting one unit of demand for sub- For the stochastic problem theoretical 10.10.1995. service j internally is qj, the cost of buy- results show that there is little hope of ing the same servicing externally. That finding approximations that always give 2 Barr, W J, Boyd, T, Inoue, Y. The is, we consider what we have saved our- near optimal solutions [6]. TINA initiative. IEEE Communica- selves by not purchasing all processing tions Magazine, March, 1993, 70–76. externally. Current work suggests that it is still possible to find good heuristics for the 3 Chapman, N, Montesi, S. Overall First we examine a greedy heuristic for stochastic problem. These heuristics can concepts and principles of TINA. the deterministic case. Defining give us knowledge which is possible to TINA-C, February 1995. (Document q implement as rules of thumb in a real- q¯ = j , no. TB_MDC.018_1.0_94.) j 1+r /d time system. Finding such heuristics and j j rules of thumb is the focus of future 4 De Prycker, M. Asynchronous trans- this can be viewed as a measure of profit work. fer mode : solution for broadband per capacity unit used if demand is met ISDN. NJ, Prentice Hall, 1995. completely for a subservice. The greedy heuristic for the single node case is 5 Conclusions 5 Dye, S, Stougie, L, Tomasgard, A. simple. The subservices are sorted by We identified the need for optimization Approximation algorithms and relax- decreasing q– and the node is filled in this j models in telecommunications as a con- ations for a service provision prob- order until fully meeting the demand sequence of deregulation, new competi- lem on a telecommunication network. before the next (critical) subservice vio- tion and new technology. Because of a Trondheim, Norwegian university of lates the node’s capacity. The heuristic
72 Telektronikk 3/4.1998 science and technology, Department 16 Onvural, R. Asynchronous transfer 19 Tomasgard, A et al. Modelling of industrial economics and technol- mode networks : performance issues. aspects of distributed processing in ogy management, 1998. (Working Boston, Artech House, 1994. telecommunication networks. Annals paper #2-98.) of Operations Research, 82, 1998, 17 Rowbotham, T. The TINA Consor- 161–184. 6 Dye, S, Stougie, L, Tomasgard, A. tium : a collaborative way forward. The stochastic single node service In: TELECOM 95, Geneva, 1995. 20 Wallace, S W. The role of uncer- provision problem. Trondheim, Nor- tainty in strategic planning. Telek- wegian university of science and 18 Tomasgard, A et al. Stochastic opti- tronikk, 94, (3), 1998, 21–23. (This technology, Department of industrial mization models for distributed com- issue.) economics and technology manage- munication networks. Trondheim, ment, 1998. (Working paper #3-98.) Norwegian university of science and technology, Department of industrial 7 Dye, S, Stougie, L, Tomasgard, A. economics and technology manage- Single node service provision with ment, 1997. (Working paper #3-97.) fixed charges. Trondheim, Nor- wegian university of science and technology, Department of industrial economics and technology manage- ment, 1998. (Working paper #4-98.)
8 Dye, S, Tomasgard, A, Wallace, S W. Feasibility in transportation net- works with supply eating arcs. Net- works, 31, 1998, 165–176.
9 Dye, S, Tomasgard, A, Wallace, S W. The stochastic service provision Shane Dye (29) graduated as B.Sc. in mathematics from problem : Benders decomposition. Canterbury University, New Zealand in 1991 and finished 1998. (In preparation.) his doctoral thesis at Massey University, New Zealand in 1994. After holding a Post Doctorate at the Dept. of indu- strial economics and technology management at the Nor- 10 Halsall, F. Data communications, wegian University of Science and Technology from 1995 to computer networks and open sys- 1997, he now holds a position at the Dept. of Management, tems. New York, Addison-Wesley, University of Canterbury. 1992. e-mail: 11 Kall, P, Wallace, S W. Stochastic programming. Chichester, Wiley, 1994.
12 Kreidi, A, Sansó, B. Optimization Asgeir Tomasgard (28) graduated from NTNU in 1993 and of a geographically distributed air- is currently studying for his Ph.D. at the Dept. of industrial ground airline telecommunication economics and technology management at NTNU with a system. Technical report, GERAD thesis entitled “Aspects of service provision and distributed publication G-94-47, 1994. processing in a telecommunications network”. This year he has been working part-time at SINTEF Industrial Manage- 13 Mullender, S (ed.) Distributed sys- ment, Economics and Logistics, where he will start as full- time researcher in December 1998. tems. New York, Addison-Wesley, 1993. e-mail: [email protected]
14 Nemhauser, G L, Wolsey, L A. Inte- ger and combinatorial optimization. New York, Wiley, 1988. (Wiley- Interscience series in discrete mathe- Stein W. Wallace (42) has been Prof. of Operations Re- matics and optimization.) search at NTNU since 1990 after spending 7 years as Senior Scientist at Chr. Michelsen Institute, Bergen, and 15 Nilsson, G, Dupuy, F, Chapman, M. 3 years at Haugesund Maritime College. Throughout, his An overview of the telecommunica- focus has been on theoretical and applied aspects of tions information networking archi- decision making under uncertainty, with applications in tecture. In: TINA 95 Conference, fisheries management, petroleum field development, Melbourne, P1, 1995. portfolio management and telecommunications. e-mail: [email protected]
Telektronikk 3/4.1998 73 How to Survive in the Future
JAN A. AUDESTAD
1 Introduction One key element in evolution of com- This paper is mainly based on two ex- panies is to prepare the road to the future. ternal sources: the definition and analysis A company lives on its portfolio of prod- This is certainly not simple in telecom- of competencies [2]; and ways in which ucts. If its portfolio is attractive to the munications as we saw in [1] – the prob- to manage innovation [3]. Important market, the company may live well and lem being the rapid and complex evolu- material is also taken from [4], [5] and be successful. However, societies change tion of the area. This complexity exists [6]. Some of these ideas were used in and the market with them. The current in all parts of the industry: technology, Telenor to determine a strategy for how portfolio may then be without value in market, competition, organisation and to assess both current assets of the com- a few years. interaction with society. pany and which competencies are re- quired to navigate safely into the future.
Core competencies 2 Categories of Assets
A company may possess assets as shown in Figure 1.
Innovation The six categories are Infrastructure • research • Innovations, manifested by research, • enabling technologies • portfolio portfolio of products, patents, sharing • IT systems • patents of knowledge, intellect and processes; • sharing of knowledge • support systems company • Historic factors such as customer base, • organisation • intellect market segmentation and brand recog- • processes nition; • People defined by their knowledge, commitment to the company and their Investments Historic factors mobility;
• production facilities People • customer base • Investments defined in broad terms as • alliances and cooperation • market segmentation investments in production facilities, • knowledge • risk management • brand recognition alliances and co-operation with similar • commitment or other companies, and risk manage- • mobility ment; • Infrastructure consisting of technolo- Figure 1 Assets gies required to produce the products, IT systems, support systems and the organisation of the company and its External forces different parts; • Core competencies which are the bun- dle of technologies, skills and intellect that will ensure success in the future. Core competencies Innovation In order to determine the strategy the company needs to consider all these ele- Internal ments. It is likewise important to assess changes continuously how well these elements are mastered and whether the set is being Infrastructure Historic factors changed as a result of external forces or company by evolution of the company itself. This is illustrated in Figure 2. External forces and internal evolution are not indepen- dent since internal changes may be a result of external forces.
Investments People The forces shaping the company are thus not static but change with time. This makes the management of these forces so difficult.
Attempts at innovating are made contin- uously in a company. The innovation is there to increase and renew the portfolio Figure 2 The company is not a stable unit
74 Telektronikk 3/4.1998 of products, improve the processes of influence may be short; in other cases it 3 System understanding, or the ‘know manufacturing, promotion and sale, and may be long. For example, the way we why’: the understanding of the differentiate the company and make it segment the market today is a function relationships between key variables; unique in the marketplace. The bigger of how we did this historically and it is 4 Motivated creativity, or the ‘care and more rapid the changes in the field changed only according to the response why’: the capacity to interrelate disci- are, the more innovative the company fed back by the market; in this case the plines and create new effects; must be in order not to lag behind and change may depend much on historic lose the competition. This means that factors. On the other hand, the change in 5 Synthesis and trained intuition, or the companies in rapidly changing industries, the customer base is less dependent on ‘perceive how and why’: the capability such as telecommunications, must keep a history (and in this way is more similar to understand or predict relationships high level of research and keep a staff of to an initial value problem in differential that are not directly measurable (see high intellect. By intellect we mean here equation) but more on what happens just also [7] where this aspect is addressed the capacity to create and use knowledge now: increase in competition, decline or in depth). [3]; it is not mere skills required to pro- improvement in customer care, market- duce and manage. In a rapidly develop- ing of replacement products, and so on. One example of this ladder of knowledge ing industry the way in which knowledge In mathematics, it is much easier to is information technology. The ‘know is distributed and shared is becoming understand and solve initial value prob- what’ can be how to use Microsoft Office very important for success. Knowledge lems than problems where the solution products to create texts and drawings; the must be shared in order to reach a critical depends on a string of historic events. ‘know how’ can be the skill to install, mass where it becomes useful. Knowl- The same can probably be said for stra- manage and maintain such systems in a edge is shared by education, co-opera- tegic problems: they are easier to handle corporate network; the ‘know why’ can tion, conferences, publishing, networks if the outcome only depends on what we be knowledge and understanding of how of acquaintances and so on. This is par- do now and not on what we did yesterday the corporate network interacts with the ticularly difficult in telecommunications and the day before. Internet and is part of the Web; the ‘care because of the combination of new and why’ can be the understanding of how immature competition and rapid evolu- One example illustrating the complexity this system is used for electronic trade, tion. This has set back joint evolution by of temporal evolution is given in [5] telecommuting and co-operative work; several years and it will take time before where the causes of the downfall of the ‘perceive how and why’ may be how more fruitful co-operation within the Encyclopaedia Britannica since 1990 are to implement such a system over many global telecommunications industry may used as a good example. Historically the actors for jointly delivering services to be re-established. In a round table debate Britannica had a large customer base, but the intelligent building: comfort, security, of ITS98 it was pointed out that much had not understood how the customer payment, encyclopaedia, dictionary ser- money is being wasted on research pro- base was built up and segmented, and vices, pay-per-view television, banking, grammes resulting at best in small inno- which external forces had formed it. and so on. vations with minor economic importance. When these forces changed – because of This seems to be a world-wide phe- the availability of cheap information on In organisations producing advanced and nomenon, and it is assumed that a com- the Internet – much of the customer base complex products, the importance of the pany can take leadership in the field only collapsed since one of the important vari- knowledge increases at least exponen- if it is able to lift its research from short ables in the ‘customer base function’ was tially down the list; that is, towards moti- term to real long term. that parents did not buy the Britannica vate creativity and synthesis. On the because of its intellectual value but to other hand, it is exponentially more Some knowledge is legally protected by do the right thing for their children. The expensive to educate people in order to patents though this protection is usually variable ‘do the right thing’ changed give them the right advanced skills and short term in the telecommunications from buying an intellectual product to cognitive knowledge (the ‘know what’ industry because the products protected buying computers with Internet connec- and ‘know how’ end of the scale). It is, may be of value only for a limited time; tions where apparently equivalent intel- therefore, cheap to create an organisation the patents are in many cases easy to lectual products were available together with much intellect but expensive to train circumvent often resulting in better prod- with many other ‘right things’. people to do the day-to-day tasks. There- ucts, and they may mask the need for fore, it is common to neglect the higher innovations and set back the evolution Organisations need knowledgeable forms of knowledge and focus the com- of new, competitive products. people; that is, people who cannot only pany on developing the concrete skills. perform the tasks the organisation re- Such organisations may be competitive What we do now soon becomes history. quires in order to produce the products now but not in the future because they Every mistake or success of the present and sell them in the market, but also lack the intellect required for change. We may subtract from or add to the image of develop the organisation into the future. see this often in job advertisements. Most the company and shape the brand recog- In [3] knowledge is divided into five companies want people with specific nition of the future. The image of the areas of increasing importance: skills to do specified tasks (mastering of company formed at present is therefore database X, programming language Y 1 Cognitive knowledge, or the ‘know probably the most important factor shap- and operating system Z). What happens what’: the rules and facts of a disci- ing the future. The history of the com- when these skills have to be replaced by pline; pany is a function of time in the sense new skills because the portfolio has that every future value is not only deter- 2 Advanced skills, or the ‘know how’: changed? It is very rare that a company mined by the value of today, but also by the ability to perform a task suffi- simply asks for clever people! It is also earlier values. In some cases the historic ciently well; well known that talent attracts talent: if
Telektronikk 3/4.1998 75 an industry is the best in its field it is assess. As we saw in [1], such extreme 3 Methodology for easy for them to get the best people. The complexity is indeed present in telecom- Determining a cost and value in investing in people are munications, and this complexity is new also studied in [8]. The findings are that so that little can be learned from history. Company’s Assets intellect is cheap; that skills are often val- I have not found any other industry in ued too high and awarded so as to create modern time changing so rapidly over One way of assessing the value of a com- wrong incentives; and that the companies such a short time as telecommunications. pany is as shown in Figure 3 [2]. with much intellectual skills are among those rated highest in the stock market. Much of the success of modern organisa- The formula given in Figure 3 is simple. tions depends on the infrastructure of the It states that: Ideally, a company should invest in prod- organisation. This infrastructure consists 1 The current value of a company is ucts that give the highest revenue. This of technologies enabling the production directly given by its portfolio of prod- is, of course, not possible and the com- of the goods, IT systems and other sup- ucts; pany has to select a set of products giv- port systems, and the way in which the ing a reasonable income over time. This company is organised. Telecommunica- 2 The future value of a company also means that the portfolio has to tions industries are facing a special prob- depends on the portfolio and the change over time. In some industries, lem in this respect: much of the enabling way it is developed into the future like telecommunications, the change is and support technologies are the products (the future potential). The future rapid and difficult to assess; in others it is created by these technologies; that is, the potential is determined by the com- slow and predictable. In order to operate products made by the industry are the petencies. in a rapidly changing environment the enabling technologies required to pro- company may need to form alliances duce the same product. Some examples In the formula we have included the with other companies in the same sector are: decentralisation of a telecommunica- support technologies which may be dif- or create new businesses by forming tions company requires telecommunica- ferent for different products and compe- alliances with companies in other sectors. tions products marketed as co-operative tencies. This is in line with [3]. The business will change over time, and work and telecommuting; the telecom- so will the alliances; alliances will some- munications operator needs a business The current value is high if the portfolio times also be formed with competitors communication network, or intranet, in has high market penetration and creates a and non-co-operative games may be the same way as other organisations; it reasonable margin. The future value is replaced by co-operative games, or a uses its own products such as networks high if the company has the compe- mixture of co-operative and non-co-oper- to create other products called services; tencies required to keep a high position ative games. In such environments, risk and it builds into these products IT com- in the market also in the future. In other management will be a key factor. Pro- ponents used as support systems for words, the position in the market is high bably, new methods need to be taken creating these products. This is different in the future only if the competencies are into use since the risks and parameters from other companies producing support such that the future portfolio will become determining them are so complex to systems: a company producing 3D draw- the leading one in the marketplace. ing and analysis software sells these products to industries using them to pro- The current value may be valid over a duce cars, ships and aircraft. On the other period of time where there is reasonably hand, oil companies use their own gaso- little uncertainty in the estimates of the line for their cars distributing this fluid; market. In telecommunications this time building constructors use buildings they may be several years in certain market have built themselves. However, here we segments and for certain products (tele- are talking of products that are much phony in the residential market), but very simpler than telecommunications prod- short (one year or less) in other segments ucts. (for example, segments related to the Internet). The changes we have seen This leads to an unusually high degree of recently in the telecommunications mar- internal trade and co-operation between ket indicate that the time where we may Current +=Future Future value potential value business units in a telecommunications assess the current value to be reasonably company. The co-operation is also quali- stable is about one to two years. tatively difficult to manage for the rea- Determined by Determined sons just mentioned. This management is On the other hand, the future potential competencies by portfolio not made simpler since the telecommuni- must be estimated over a long time in cations company is not a Porterian chain order to identify the really important but rather a value network [4], favouring competencies. Note that one such com- different ways of co-operating internally petency may just be the ability to make Products Competencies and externally. assessments in an environment with much uncertainty. Reasonable planning Support The final force shaping the company is horizons for the future potential may then technology the core competencies. We will soon be five or more years: The horizon come back to these. should not be too close because then the changes may not be big enough. To iden- Figure 3 The future value tify some critical competency, the hori-
76 Telektronikk 3/4.1998 zon should not be too remote because ucts that the company should protect or High then important competencies may be capture. Product 4 has small margins and blurred by excessive uncertainty. much growth. This product is then likely Product 1 Product 3 to become a problem for the company in We have also included the support tech- the future unless its profitability is im- nology in the formula. This element was proved. The strategy for such products is not included in [2] but its importance is to change their profitability or, if that is indeed recognised in [3]. The support not possible, withdraw them from the Margin technologies are required both for creat- market. ing the product portfolio and for support- Product 2 Product 4 ing the competencies. IT systems are the The products in this diagram are not basis for products like electronic trade static. Over time they may move from and electronic newspapers, and telecom- one category to another; for example, a munications technologies are required for product of type 3 may become a cash distribution and remote control of such cow or become unprofitable. This Low products. Competencies like customer dynamics must be built into the model in (or negative) care, sales management, logistics and order to prepare the company changes Low Growth High customer base management all rely on in the portfolio. (or negative) IT systems. The vulnerability may be assessed by use market of a price differentiation diagram as share 4 Current Value Analysis shown in Figure 5. The current value of a company may be Functionality may replace price and vice characterised by two types of diagrams: versa. This is illustrated in the diagram Figure 4 Margin vs. growth one diagram relating the margin and the by splitting it into two regions, the blue growth potential of all the products of the region and the red region. In the blue company; and one diagram showing region the company in question has ad- price and differentiation potential for vantage either because the price is low each product. The first diagram is shown and the product is well differentiated product is superior from a differentiation in Figure 4. (product 1); the price is so low that it point of view and the price is not so high compensates for the superior function- that it masks the differentiation advan- The growth parameter may be the esti- ality of the competitor (product 2); the tage (product 3). For a product in the red mated growth rate at present. The esti- mate may be based on the average taken over some years (the compound annual growth rate). The other axis shows the margin. Each product may be visualised as disks with a size proportional to the overall revenue of that product in the High market. The market share of the company under consideration and its competitors may be shown as sectors. Product 4
Figure 4 shows four products. Product 1 Product 5 has a high margin but low (or even nega- tive) growth potential. This is a typical cash cow that should be exploited now, Product 3 but small investments should be allocated Relative in this area since the product is likely to price disappear. Product 2 has a low or nega- (own price/ tive margin or shows little or negative competitor price) Product 6 growth. Such products do not contribute to the wealth of the company, neither now nor in the future, and should be ter- Product 1 minated. The problem with such a prod- uct is that the termination cost may be high, or that the company is obliged to Product 2 keep it for fulfilment of licensing re- quirements. Low Own Comparable Competition Product 3 offers high margins and much advantage fuctionality advantage growth potential, and is thus attractive from a business aspect. These are prod- Figure 5 Vulnerability
Telektronikk 3/4.1998 77 area the following applies: the price is • Financial measures; The ideas behind the theory are: high and the differentiation advantage is • Customer satisfaction and market share; 1 Almost all systems will develop along not sufficient to compensate for it (prod- the most natural path and almost all uct 4); and the product has neither • Efficiency of operational processes; shapes in nature will take the most nat- enough differentiation advantage nor • Organisational learning and assess- ural form from a mathematical point of price advantage (products 5 and 6). ment of the knowledge base of the view; company. 2 A generic or canonical description 5 Future Value: exists for such forms; Competencies It is important that other measures than financial are considered because it will 3 Many such forms are structurally normally be found that the core compe- stable; that is, they are not altered The competencies are the collection of tencies are related to the mastering of by small changes; skills and knowledge that enable the these measures ([2], [3]) rather than the company to succeed in the market. One 4 This generic view also applies to the traditional financial measures. As is eas- systematic way of determining the com- properties of the space in which sys- ily observed in the market, the financial petencies is a series of investigations of tems develop. status of a company may change so fast the following items: that it is impossible to build a long term Catastrophes are classified into six fami- • The vision of the company; strategy upon it. lies of elementary forms (Thom’s classi- • The key success factors; fication theorem). Non-elementary and The complex dynamics of telecommuni- more complex forms exist, but they may • The industry drivers; cations is driven by a large set of param- not be structurally stable. To us, the most eters called the industry drivers. These • Customers and customer interfaces; interesting of these families is the two- drivers are external and one reason for parameter cusp catastrophe, since several • The benefits provided to the cus- defining them is that the company needs of the analyses we have done in Telenor tomers; to define responses to these drivers or indicate that the future is determined by adapt to them as required. The drivers are • Identification of competencies; two aggregated, free variables (for ex- then the same as we defined as external ample, market growth and regulation) • Consolidating and ranking the com- forces in Figure 2. The responses are the and one dependent variable (margin). petencies; internal changes. Though there are many detailed vari- • Determining the company’s mastering ables, they tend to aggregate into a few The key drivers may be grouped in of the competencies; variables describing the system in aggregated sets as for example: enough detail. Similar behaviour is seen • Propose how holes in the competencies • Market demand variables comprising in the development of fish stocks [9], can be filled; the impact of national or international where the external variables are the • Build up enthusiasm for change in the economic conditions, integration with efficiency of the fishing method and the organisation; products of other industries, identifica- number of fishing vessels. If we take, for tion of supplementary, complementary instance, the performance of the com- • Introduce changes as required in the or replacement products, and demo- pany as the dependent variable and the organisation. graphic changes; two parameters (or free variables) market growth and regulation as independent Background information may be ob- • Technological variables such as speed variables, the performance of the com- tained by use of workshops where execu- of introduction of new technologies, pany as a function of time can be viewed tives and key personnel of the company technical agility of competitors, invest- as a trajectory on a two-dimensional sur- participate to create a common view of ment requirements and capital binding, face as shown in Figure 6. the company, the industry sector to and the impact of unexpected techno- which it belongs, and the forces that will logical evolution; According to Thom’s classification theo- shape the industry. The process is then as • Regulatory variables including pricing rem the most natural form of this type of follows. constraints, co-operation requirements, surface will contain a cusp singularity as and delivery obligations in non-profit- shown in Figure 6 (the argument for this First, the long range goals of the com- able markets. conclusion in too long and complex to be pany are defined and compared with the reproduced here). The behaviour of the vision formulated for the company. In The impact of the changes is related to company may be as indicated by the two this process it may be found that even and may be understood in terms of cata- trajectories. In some cases the evolution the vision should be adjusted, for ex- strophe theory. Catastrophe theory was will be smooth (trajectory 1). If the tra- ample, in order to meet new expectations proposed by the French mathematician jectory passes over the fold (trajectory 2), of the market. To achieve the vision the Thom during the 1950s. The theory can- there will be a sudden change in perfor- company needs to see or measure which not easily be used to predict a given mance. What is characteristic for this factors are determining the success of behaviour but it may help understand case is that the changes may be small for the company. These are the key success qualitatively certain types of sudden a large range of variation of the two factors. It is important that these factors changes which may occur. The mathe- parameters (in the smooth areas to the can be measured in order to assess how matics and several applications of cata- right or left of the fold) so that it is diffi- well the company is able to fulfil the strophe theory are found in [9] and [10]. cult to foresee the catastrophe from vision. These factors may be: merely observing the performance. If the
78 Telektronikk 3/4.1998 trajectory is reversed, the performance performance will change as shown by the dotted line. The existence of the hystheresies makes the performance irreversible. This is exactly the behaviour observed for the fisheries in the North Atlantic Ocean. It may also explain why a company may prosper one day and be bankrupt the next day.
However, as was said above, this is only a qualitative description of the behaviour. It will probably be impossible to both find mild the catastrophe surface and the trajectory severe describing the performance of the com- pany. The importance lies in that this de- little regulation scription will make management and oth- ers aware of this type of non-linear be- cusp Trajectory 1 haviour and build it into the strategy plans.
The company needs to define their cus- tomers accurately. In the telecommuni- cations industry a vertical company like catastrophy fold Telenor will have several types of cus- much hystheresies Trajectory 2 tomers requiring different management:
• Residential customers who may be market growth divided further into segments requiring different handling; • Business customers requiring tele- Figure 6 Performance trajectories communications as a primary product; • Business customers using telecommu- nications as production factors in their own primary products (electronic commerce, bank transactions and dis- Key source of tance learning); differentiating Important competencies Core competencies • Other network operators requiring interconnect; Product Consultative • Service providers requiring infrastruc- sourcing selling ture services in order to complete their product. Intellectual skills For each customer group the company should analyse in depth what the group requires and how well this requirement Potential for is met. differentiating
From these analyses it may now be pos- sible to identify which competencies the company requires in order to progress Basic Technology B safely into the future. In this way a con- competencies Technology A siderable set of competencies may be defined. The competencies may then be ranked as shown in Figure 7. Customer Logistics The example is not taken from any exist- care ing business or company; it is included only for visualisation of the ranking. The Small potential competencies may be ranked along two for differentiating axes: the abscissa indicates how im- Baseline Enabling leadership Shaping industry portant the competency is for enabling skills leadership and the ordinate is assessing Figure 7 Classification of competencies
Telektronikk 3/4.1998 79 its potential for differentiation. Some endurance of a company’s leadership in The core competencies must be such that competencies are just baseline skills the market. they result in values to customers in the (logistics in the example) which the com- long run. It is argued that one core com- pany must possess but has no impact on We saw that the core competencies are petency of every company should be how the industry will evolve in the those competencies which differentiate related to understanding or serving the future. On the other hand, there may be the company from its competitors and is customers. Several computer firms with competencies (such as technology B and the reason why its products may be pre- the world’s best technology in its niche, intellectual skills) which will shape the ferred by the customers. At the same have failed in just this area because they future. Between these extremes there time, it is the competencies which will have either had difficulties serving the may be other competencies which may form the future of the industry. market (Cray Research) or adapting to be more than basic skills but with only changes in the market (Norsk Data). limited impact on the future evolution. In addition, the core competencies are difficult to copy by competitors either The value of core competencies may On the ordinate we show the compe- because they are expensive to create, become even more important when an tency’s potential for differentiating one difficult to build up, or difficult to under- industry is disaggregated. This has company from another. In the example, stand. happened in the biotechnology industry logistics is regarded to have a small by coalition of several specialised com- impact on the potential for differentia- Successful companies target few core panies in order to create an easily man- tion. Product sourcing and consultative competencies – often only two or three, oeuvrable aggregate industry, where selling are, on the other hand, compe- and seldom more than five. The core each element may develop along their tencies which are key sources of differ- competencies are not the traditional func- own core competencies. This has also entiation. tions such as finance, sales, production or been the case for a long time for the engineering, though these are the activi- watch industry in Switzerland and the The diagram may be divided into three ties around which organisations were furniture industry in Norway. Here there parts as shown: basic competencies, formed in the past [3]. Instead, they are are companies specialised in making important competencies and core com- multifunctional activities cutting across parts and being so good at doing this that petencies. This division is important in the organisation not favouring power- no other companies can match them on order to identify where the company building by product or management neither quality nor price. Now the same should focus its strategy. As part of this groups trying to gain dominance of the evolution is coming in telecommunica- analysis the company should assess how company. In this way, the core compe- tions where the regulation of the libe- well the different competencies are mas- tencies are not products but rather intel- ralised market has first of all prepared tered in order to identify where resources lectual competencies that can sustain a the ground for companies operating in should be placed for improved competi- leading edge even if some products narrow and profitable niches [1]. tion and ‘world mastery’. Building basic wither and disappear or are replaced by Telecommunications, being an industry and important competencies will be nor- others. Such competencies may then be of value networks, is a good candidate mal educational activities in the com- based on skills in technologies, methods, for decomposition of the industry into pany. Building core competencies will processes, market assessment and so on, niches which may lead to new types of be a strategic endeavour. where each such competency does not co-operative behaviour in the market. support only one but several products. The development is still too new to Competencies can be obtained in differ- assess the result of such decomposition. ent ways. Some competencies may be One such core competency may, for home-grown; that is, developed by the example, be the management of canni- The new development called C4 con- company itself by funding research and balising own products [11]. Such canni- vergence, where the four Cs stand for educating employees. Competencies can balisation may be the most important Content, Computer, Communication and also be obtained by alliances where com- incentive for radical innovations. It is Consumer electronics, may give rise to petencies the company lacks are obtained difficult not because it creates new busi- much more co-operation and competi- from the partners. There are also cases nesses but because it may remove part of tion. One core competency may then be where a competency is present in part of the business. In an industry where the to discover, establish and maintain co- the company. In such cases the compe- changes are fast, the understanding and operation in such markets. This core tency may be spread to other parts of the management of cannibalisation may be competency may be supported by the company by leveraging the competency. the most important core competency. competencies of bundling products and Finally, competencies may be bought or Hewlett Packard is one example of a technologies, binding customers through hired by buying a company (for example company where this competency is well part of the portfolio contained in the bun- in logistics) or hiring individuals or con- developed. dle, stimulating buyers by letting some sultants possessing the competency. part of the offer being free of charge (as An example of the utilisation of technical for some Internet products), and create a competency and marketing competency shared market recognition which is not 6 Characteristics of Core is Honda [3]. Honda is a leading com- associated with a single product but with Competencies pany in small engine design and inno- the synergy obtained by the combination. vative marketing and distribution. These It is the core competencies that will make competencies combined provide the the company successful in the future. company with a core competency which They are the main factors that will ensure has been difficult for other companies to match.
80 Telektronikk 3/4.1998 7 Conclusion References 9 Casti, J. Alternate realities : mathe- matical models of nature and man. The main concerns of a company in the 1 Audestad, J A. Telecommunications Wiley, 1989. telecommunications market should now and complexity. Telektronikk, 94, be: (3/4), 1998, 2–20. (This issue.) 10 Poston, T, Stewart, I. Catastrophe theory and its applications. Pitman, • Understand the highly dynamic 2 Prahalad, C K, Hammel, G. The core 1978. changes now taking place in tech- competencies of the corporation. nologies, markets, external regulations Harvard Business Review, May-June, 11 Chandy, R K, Tellis, G J. Organizing and competition, and recognise that 1990. for radical product innovation : the survival in the field will require excel- overlooked role of willingness to lence in strategic awareness and man- 3 Quinn, J B, Baruch, J J, Zien, K A. cannibalize. Journal of Marketing oeuvrability; Innovation explosion. The Free Press, Research, 35, 1998, 474–487. • Increase the intellectual value of the 1997. company by recognising that the cost of the company is linked to the de- 4 Stabell, C B, Fjeldstad, Ø D. Con- tailed ‘know what’ and ‘know how’ figuring value for competitive ad- skills of the company, while the future vantage : on chains, shops and net- success depends on how well the com- works. Strategic Management Jour- pany is organised around intellect (the nal, 19, 1998, 413–437. ‘care why’ and ‘perceive how and why’ knowledge) noting that it is 5 Evans, P B, Wuster, T S. Strategy cheap from a financial viewpoint but and the new economics of invest- difficult from a management viewpoint ments. Harvard Business Review, to retain the best intellects of the Sept-Oct, 1997. organisation; 6 Løwendahl, B, Revang, Ø. Chal- • Keep an accurate overview of the port- lenges to existing strategy theory in folio of products with regard to sus- a post-industrial society. Strategic tainability, margin created by each Management Journal, 19, 1998, product and vulnerability to compe- 755–773. tition. This is a dynamic process re- quiring regular analysis in order to 7 de Bono, E. The use of lateral think- detect changes which may require ing. Pelican, 1971. strategic decisions; • Identify and create core competencies 8 Pfeffer, J. Six dangerous myths about which can ensure a safe journey into pay. Harvard Business Review, May- the future. June, 1998.
Jan A. Audestad (56) is Senior Advisor for the Corporate Management of Telenor. He is also Adjunct Professor of telematics at the Norwegian University of Science and Technology. e-mail: [email protected]
Telektronikk 3/4.1998 81
82 Telektronikk 3/4.1998 Special
Telektronikk 3/4.1998 83 84 Telektronikk 3/4.1998 Voice quality under ATM cell loss
JENS BERGER AND MECHTHILD STOER
This study tries to assess the effect of ATM cell loss and into larger packets, i.e. when the loss of one cell may lead to the AAL packet loss on voice connections. To this aim, cell loss of a larger packet. We assume that packet sizes have an losses on a 64 kbit/s CBR stream through one ATM link impact on voice quality also for voice over IP. It also turns out were measured under high load, and the resulting cell loss that PSQM+ and DT-SQE values show a high correlation with trace merged off-line with recorded speech files, simulating the outcome of the auditory test. This gives hope for methods AAL1 and AAL5 protocols. We study the impact of differ- for instrumental voice quality evaluation being applicable for ent cell stuffing techniques and of packet loss for packets of studies and laboratory experiments. different lengths. Delay and delay variations have not been measured nor introduced into the speech files. Voice quality The impact of ATM cell loss on voice quality has also been was assessed by a formal auditory (‘subjective’) test and by studied by Meky and Saadawi [12]. They used artificial distri- several instrumental (‘objective’) methods: PSQM, PSQM+ butions (uniform, Poisson, etc.) for the cell loss process, and a and DT-SQE. This gives rise to comparisons between cell specially designed instrumental method to evaluate the speech loss rates, auditory evaluation and instrumental estimation quality. This method is also reported to have good correlation of speech quality. with subjective evaluation. This work was financed by EURESCOM Project P603 and In our paper we used, instead of artificial cell loss distributions, by Telenor Research Project KN0401. actually measured cell losses in an ATM switch with high back- ground load and small buffer. Cell losses do not occur evenly 1 Introduction spread but in bursts. Since the instrumental methods were de- signed for evenly spread quality degradations, it was not clear The background for this study was the selection and definition from the beginning that they would also estimate ‘correctly’ of important Quality of Service parameters and the development quality degradations occurring in bursts. of measurement methods. This was the scope of EURESCOM project P603 ‘Quality of Service: Measurement Method Selec- A more detailed version of this paper is found in [5]. tion’, see [3] and [4]. It was found that voice quality still is an important parameter, especially with emerging new techno- Section 2 describes how speech files impaired by cell loss were logies for transfer of voice and data. The methods chosen to produced. Section 3 describes the evaluation methods consisting measure voice quality were: of the auditory test and the instrumental methods, and sections 4 and 5 contain results and conclusions. • A formal auditory test following ITU-T recommendation P.800 [9] • The instrumental methods 2 Measurement set-up and test - ‘Perceptual Speech Quality Measure’ (PSQM) recom- conditions mended for codec evaluation in ITU-T P.861 [11] An ATM cell stream simulating a 64 kbit/s speech connection - PSQM+ (a modification of PSQM to assess bit error condi- was fed together with a background load into an ATM switch. tions) The switch is a special designed measurement and experiment switch [13]. The switch was configured so cell loss occurred in - Deutsche Telekom – Speech Quality Estimation (DT-SQE) a buffer of length 70 toward a 155.52 Mbit/s link. The back- [2] designed to assess codecs. ground load came from synthetic traffic generators (STG) [6]. For a given background load, we registered over a certain time All in all this investigation contained more than 50 different the lost cells of the primary stream. Segments of the cell loss configurations for speech transmission: trace were then merged with one of several digital speech files. • Voice transmission via ‘real’ telephone networks (ISDN, Thus we produced several impaired speech files. Some more POTS, GSM) details on the primary 64 kbit/s stream, the background traffic, and the impairment process follow. • Transmission via simulated connections
- Codecs 2.1 Description of the primary stream - ATM connections At Telenor Research, we had at hand an automatic traffic gener- - Background noises ator (Alcatel 8643 ATM Traffic Generator Analyzer, ATGA, • Reference system MNRU (see ITU-T P.810 [10]). see [1]) that is able to generate ATM cells at constant bit rate, with a given virtual channel and virtual path number, with In this paper, we concentrate on the application of these sequence numbering in the payload, and otherwise empty pay- methods to ATM cell loss. This gives rise to interesting com- load. We used these features to emulate a 64 kbit/s user data parisons between cell loss, auditory and instrumental speech stream representing the speech connection. AAL1 is intended quality evaluations. for transfer of voice connections, see [14]. This protocol re- serves 47 bytes of the ATM cell’s payload for user data. In the It turns out that up to 2.5 % cell loss inside active speech inter- ATGA’s cell generation program we set the user’s payload to vals still produces fair quality, but that the same cell loss rate 47 bytes and the rate to 64 kbit/s. This corresponds to a rate of leads to intolerable quality degradation when cells are packed 170.21 cells per second.
Telektronikk 3/4.1998 85 Table 1 Cell loss traces non-overlapping windows of 8 seconds, which corresponds to 1362 sent cells of the primary stream. trace name A C E We simulated packing of a speech file into 1362 cells (with -3 -3 -3 cell loss on 1.125 * 10 1.097 * 10 0.879 * 10 sample 1 into cell 1), where 47 samples would fit into 1 cell, as primary stream the AAL1 protocol would do. There is one difficulty here: The cell payload is 47 bytes, but 47 samples of the original speech measuring time 78 70 74 files consist of twice 47 bytes. A more proper simulation pro- (min) cedure would have been to (sample-wise) code the speech files scenario HQTV- CBR- HSD- using A- or µ-law, thus halving the file size, then to simulate dominated dominated dominated the packing of coded samples into cells, and to decode the files again. We skipped the coding/decoding procedure, since quality reduction by A- or µ-law coding is negligible.
There are several ways to replace samples in lost cells: 1. Silence insertion: The lost 47 samples are replaced by 47 2.2 Description of the background traffic samples of value 0, which denotes silence in linear PCM. 2. Repeating: The lost 47 samples are replaced by the 47 For generation of the background traffic we used the Synthetic samples in the last received cell. If the first cell was lost, Traffic Generator (STG) developed by Wandel & Golterman, its samples are replaced by silence. Telenor Research, and SINTEF Tele&Data. A high-level de- scription of its possibilities and its interface can be found in [6]. 3. Packet replacement: Here we simulate the sending of the file Source types can be specified inside the STG as Markov models in packets of k cells. Whenever one or more cells are lost in with state sojourn times, state transition probabilities, and states a packet, the whole packet is lost. This simulates the AAL5 corresponding to the sending of a predefined cell pattern. We protocol for sending of packets. Samples in lost packets are used three types of sources: constant bit rate traffic (CBR), high replaced by silence-value 0. We used the arbitrary values of speed data traffic (HSD), and high-quality television (HQTV), 3 and 6 for k. whose Markov source models were defined in [7]. From these source types three traffic scenarios are set together: To be exact, the user payload under the AAL5 protocol is on the average between 47 and 48 bytes, but we ignored the exact • An HQTV-dominated traffic scenario where the load from the values. sources of the three types (HQTV, HSD, CBR) contributes approximately in the ratio 0.6 : 0.2 : 0.2; The three replacement methods are abbreviated ‘s’ for silence • An HSD-dominated traffic scenario where the load from the replacement, ‘r’ for repeating, and ‘s3’ or ‘s6’ for replacement sources of the three types (HQTV, HSD, CBR) contributes of packets of 3, resp. 6, cells by silence. approximately in the ratio 0.2 : 0.6 : 0.2; • A CBR-dominated traffic scenario where the load from the 2.4 Conditions tested sources of the three types (HQTV, HSD, CBR) contributes approximately in the ratio 0.2 : 0.2 : 0.6. For the auditory test, five German language files of 32 seconds length each were prepared, which were to be impaired under The total mean load of each of these traffic mixes was set such five different conditions. The 32-second files contained four 8 that the total stationary cell loss ratio becomes approximately second segments spoken by two male and two female speakers. 10-3. This high value (instead of the more usual 10-9) was The 8 second segments each contained two whole sentences by chosen, because we want to look at the effect of the otherwise one speaker. rather seldom cell loss bursts. Another reason for this choice of cell loss ratio was that the speech file segments to be impaired An example condition in the ATM case is “each of the four 8 were maximally 8 seconds long. This corresponds to about 1362 second segments of a 32 second file is impaired by loss of x cells. packets of size k of the xxx-dominated scenario; and the loss occurs in a period of active speech.” By varying the number of Table 1 shows the cell loss rates (cells lost relative to total num- packets x and their size k, we received five conditions. ber of cells sent) on the primary stream for measuring times of about one hour, for each scenario. We did not measure the total We produced the impairments in the following way. From one cell loss rate of the primary stream together with the back- of the cell-loss traces A to E we chose four 8 second windows, ground streams. in which about x packets of size k were lost by the packet re- placement method described in Section 2.3. The loss of x 2.3 Merging the cell loss pattern with the packets usually occurred in bursts of maximal length 70 ms. speech files A difficulty was to place the burst inside a period of active speech of an 8 second segment. We tried, inside one 32 second file, to make one burst meet the end of a sentence, another the The available speech files were coded in linear PCM with beginning of a sentence, and the rest the middle of a sentence. sampling rate 8 kHz and 2 bytes per sample. They consisted of segments 8 second long. Each cell loss trace file was split into
86 Telektronikk 3/4.1998 Table 2 shows the conditions. The trace name refers to Table 1 Table 2 Impaired files for auditory test (cell loss traces), and the replacement method refers to section 2.3. impaired trace packets lost packet size replacement file in 8 sec (k cells) method These five conditions were included in a set of altogether 53 segment (x) different conditions. Therefore the results are directly compar- able with the considered references and the other conditions in g30_atm A 11–14 1 s the test. The speech quality was evaluated in an auditory test performed at Deutsche Telekom Berkom. The quality of the g31_atm A 12–14 1 r impaired speech file was also evaluated by PSQM [11], PSQM+ g32_atm C 7–8 3 s3 (a modification of PSQM) and DT-SQE [2] as a function of time. In the next section we describe the auditory test and the g33_atm C7 6 s6 instrumental methods. g34_atm E 11–12 6 s6
3 Evaluation methods g30_atm and g31_atm are using the same cell loss patterns. So are g32_atm and g33_atm. 3.1 Description of the auditory test
The best model for evaluation of speech transmission quality would be a set of normal free conversations between two test participants. Such kinds of tests make it possible to consider and Scoring to evaluate all characteristics that influence the speech quality In accordance with ITU-T P.800 [9] a five-step Absolute Cate- (or better, the capability for conversation) in an end-to-end con- gory Rating (ACR) scale is offered for scoring: nection. But conversational tests with two partners in a test lab- oratory are very time consuming and require a high number of 5 excellent test participants. Therefore, for estimation of the speech quality 4 good of transmission systems (i.e. codecs, filter systems) the usage of 3 fair Listening-Only-Tests (LOTs) is very common. These tests are 2 poor well suited for a very efficient estimation of the one-way trans- 1 bad mission quality. Some features of quality, which are only noticed in conversations, cannot influence the test results (e.g. All test participants (24 in our case) score all conditions in the signal delay, talker echoes, semi-duplex characteristics). The test. The 53 tested ‘conditions’ correspond to 53 different 32 efficiency of LOTs makes it possible to evaluate a high number second speech files. Five of these were impaired by ATM cell of test conditions in one listening experiment using the same loss, as described in the previous section, the other conditions test participants and well-prepared speech material for statistical involved ISDN, GSM, POTS connections, and different codec validity. Another advantage of LOTs is the off-line test proce- simulations. A full description can be found in [4]. dure; in the test recorded speech samples are used. There is no need to have the current telephone connection available for the After each offered speech sample (length: 8 sec) each test auditory test. listener has to score the perceived overall quality. The speech transmission quality will be described by a Mean Opinion Score Design of the auditory test (MOS) for each condition. The MOS value is here defined as the mean value of all individual results of the test listener and Test participants the four speakers (in total: 96 single results for every condition). • The group of test participants consists of typical non-tech- nical subjects selected from a pool of listeners used for audi- 3.1 Methods for instrumental assessment tory tests. • The age and the educational background of the subjects pro- Instrumental approaches to speech quality prediction try to vide a good representation of the typical telephone user popu- achieve a high correlation between the predicted quality values lation. and the scores gained by auditory tests. Most instrumental methods for speech quality estimation compare the (undistorted) • The percentage of female and male listeners is 50 % each. source speech signal and the impaired output signal of the trans- • All test participants were previously audiometrically checked mission system. for normal hearing threshold by pure tone audiometry. A great number of instrumental approaches work in several Speech material steps. The first step usually eliminates signal differences that are irrelevant in the modelled auditory test (for example total delay Each 32 second test file was split into four 8-second speech and level differences). The next stage transforms both signals to samples. The used speech samples consist of two short German an ‘internal representation’ using psycho-acoustic models for sentences coupled by a silence interval spoken by one person. the human sound perception. The spread (including multidimen- The whole 32 second file is spoken by two males and two sional aspects) between both pre-processed signals is computed females. All samples are selected from a phonologically and will be used for estimating a quality value. In this way mod- balanced speech database. ern instrumental approaches based on psycho-acoustic and cog-
Telektronikk 3/4.1998 87 input output signal signal System under test
Model of sound perception
Distance/likelihood of Weighting predicted signal characteristics function quality value SQ' τ Model of sound perception
equalisation of level and delay Figure 1 General structure of instrumental speech quality estimation approaches
nitive knowledge are models of the auditory test with human Features of the used instrumental approaches test listeners (Figure 1). • DT-SQE (Deutsche Telekom – Speech Quality Estimation) and PSQM (Perceptual Speech Quality Measure) follow this Several approaches or methods are known which differ in e.g.: common structure and components as shown in Figure 1. • The kind of signal pre-processing; • The first step in both measures is the total delay and gain • The used psycho-acoustical models; adjustment (these values are given or computed before). • The determination of a value describing the speech quality; • DT-SQE and PSQM consider the telephone band limitation and and the receiving function of an ‘ordinary’ handset. • The database for definition of the fitting function to transform • Both measures are based on a short-time spectral analysis the technical speech quality value to a predicted MOS value. (tW = 16 … 20 msec) and compute the levels in the critical bands for each signal and each segment (frequency warping). DT-SQE has been designed in the Deutsche Telekom Berkom - PSQM divides the spectrum into 0.25 Bark bands; DT-SQE research group ‘Quality and Acceptance of Voice Services’ only into 1 Bark band. (former: ‘Speech Quality Measurement, Psycho-acoustics’ Research Centre of Deutsche Telekom AG). A former approach - Additionally, DT-SQE reduces the influence of total fre- of DT-SQE was published in connection with an investigation quency shape in this step. of DCME systems in [2]. The current version of DT-SQE used - PSQM adds a Hoth noise to both signal representations. in EURESCOM project P603 was tuned using German speech samples taken from ITU-T and ETSI speech quality tests. This • Intensity warping of the critical band levels: approach will be invited in ITU named as T-OSQA. - DT-SQE is based strongly on the Zwicker model of calcu- lation of specific loudness. Spectral masking and the com- PSQM (Perceptual Speech Quality Measure) is a method for pression are taken from the Zwicker model [15]. measuring the quality of narrowband (300 – 3400 Hz) speech codecs. It is recommended by ITU-T P.861 [11]. PSQM 0.0 is a - PSQM considers masking effects only in a simplified way version of PSQM in which distortions in silence intervals are and uses a much higher compression than Zwicker. weighted with 0.0 (i.e. not considered at all). PSQM 0.2 weights • PSQM calculates a ‘loudness scaling factor’ to adjust the silence intervals with 0.2. Silence weight 0.2 is recommended loudness in the current segment. by ITU-T P.861 [11]. PSQM+ is an experimental version of PSQM designed for better handling of bit-errors in GSM. • DT-SQE reduces differences between both signal representa- tions, which have nil or a restricted influence in the auditory PSQM and DT-SQE have been developed and optimised for test. One part of this step is also a slight asymmetry process- evaluating and rating speech codecs. Only limited experience ing (see PSQM). has been gained by experts using these methods to assess end- • Both modified short time loudness patterns are used for cal- to-end quality of international (inter)connections or to assess culation of a quality value in the current segment. ATM or IP connections. • The computed similarity is the main result of measuring the speech quality in DT-SQE. The mean value of all short time results yields the internal speech quality value.
88 Telektronikk 3/4.1998 • In PSQM the result of a computed ‘cognitive subtraction’ which was optimised for the ITU-T 8 kbit Test, 1994. But in the after an ‘asymmetry processing’ describes the current quality context of this test the results should only be used as ‘quality value. describing values’, not as estimated MOS.
It is often desirable to express voice quality in MOS values, but it is difficult to determine a unique function which transforms 4 Results of the analysis the DT-SQE or the PSQM values to the estimated MOS value. In this section we compare the subjective MOS values with the This is because absolute MOS values depend on the context of instrumental values for the files impaired by ATM cell loss. the auditory tests. Also cultural differences or differences in lan- Table 3 shows the results for the 8-second segments and Table 4 guage can have significant effect on the auditory MOS. There- the results for the 32-second files composed of the 8-second fore, we decided in this study to use the internal speech quality segments. One should note that the subjective MOS in Table 3 values directly. For DT-SQE a mapping function was used,
Table 3 Results from auditory test for 8-second segments
segment subj. MOS PSQM 0.0 PSQM 0.2 PSQM+ 0.0 DT-SQE-MOS
g30_atm.1 3.292 0.079 0.075 0.083 3.959 g30_atm.2 3.125 0.528 0.301 0.593 3.924 g30_atm.3 2.792 0.869 0.570 0.993 3.726 g30_atm.4 3.042 0.410 0.266 0.477 3.970 g31_atm.1 3.667 0.218 0.158 0.215 3.829 g31_atm.2 4.292 0.113 0.090 0.114 3.964 g31_atm.3 4.667 0.213 0.132 0.206 3.975 g31_atm.4 3.458 0.695 0.470 0.697 3.779 g32_atm.1 2.750 0.211 0.169 0.872 3.487 g32_atm.2 2.083 0.618 0.389 1.201 3.777 g32_atm.3 2.708 0.451 0.340 1.188 3.303 g32_atm.4 2.750 0.413 0.322 0.779 3.918 g33_atm.1 2.250 0.371 0.242 2.088 3.080 g33_atm.2 2.208 0.381 0.282 1.460 3.063 g33_atm.3 2.708 0.646 0.485 2.992 3.152 g33_atm.4 2.625 0.519 0.397 2.685 3.239 g34_atm.1 2.083 0.535 0.367 3.923 2.809 g34_atm.2 2.417 0.645 0.396 4.656 2.234 g34_atm.3 1.792 0.951 0.557 7.399 2.392 g34_atm.4 1.833 0.967 0.693 4.666 2.699
Table 4 Results from auditory test for 32-second files
segment subj. MOS PSQM 0.0 PSQM 0.2 PSQM+ 0.0 DT-SQE-MOS
g30_atm 3.062 0.483 0.312 0.55 3.849 g31_atm 4.021 0.326 0.219 0.32 3.873 g32_atm 2.573 0.411 0.306 0.99 3.637 g33_atm 2.448 0.493 0.365 2.39 3.055 g34_atm 2.031 0.773 0.509 5.04 2.634
Telektronikk 3/4.1998 89 is averaged over 24 observations only. The 95 % confidence 4.1 Comparison of instrumental values with intervals are of size 0.24 to 0.52. So these values have to be subjective MOS treated carefully. In contrast, the subjective MOS values in Table 3 have been averaged over 96 observations. The 95 % In Figure 2 to Figure 5 the instrumental values are plotted confidence intervals for the subjective MOS values in Table 4 against the subjective MOS for each of the 8-second segments. are smaller, of size 0.18 to 0.23.
PSQM-0.0 PSQM-0.2 1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0 12345 12345 MOS MOS cell replacement, silence, g30_atm cell replacement, silence, g30_atm cell replacement, repetition, g31_atm cell replacement, repetition, g31_atm packet replacement, 3 cells, g32_atm packet replacement, 3 cells, g32_atm packet replacement, 6 cells, g33_atm packet replacement, 6 cells, g33_atm packet replacement, 6 cells, g34_atm packet replacement, 6 cells, g34_atm Figure 2 PSQM 0.0 versus subjective MOS. 8-second segments Figure 3 PSQM 0.2 versus subjective MOS. 8-second segments
PSQM+0.0 DT-SQE-MOS 8 5
4 6
3 4 2
2 1
0 0 12345 12345 MOS MOS cell replacement, silence, g30_atm cell replacement, silence, g30_atm cell replacement, repetition, g31_atm cell replacement, repetition, g31_atm packet replacement, 3 cells, g32_atm packet replacement, 3 cells, g32_atm packet replacement, 6 cells, g33_atm packet replacement, 6 cells, g33_atm packet replacement, 6 cells, g34_atm packet replacement, 6 cells, g34_atm Figure 4 PSQM+0.0 versus subjective MOS. 8-second segments Figure 5 DT-SQE versus subjective MOS. 8-second segments
90 Telektronikk 3/4.1998 Table 5 shows the correlation between the subjective MOS and Table 5 Correlation of MOS to instrumental values for 8-second segments a linear or quadratic fitting function in the respective instrumen- tal values. instrumental value linear quadratic
Even taking into account the unreliability of the subjective PSQM 0.0 0.63 0.65 MOS data for the 8-second segments, it seems that PSQM+ and DT-SQE have a better correlation with the subjective PSQM 0.2 0.64 0.68 MOS than PSQM has, no matter which silent weight is used. PSQM 0.4 0.67 0.72
It does not really make sense to compute correlations of sub- PSQM+ 0.0 0.70 0.77 jective MOS with the instrumental values for the 32-second PSQM+ 0.2 0.71 0.77 files, because there were only five files. Instead, we plot the results in Figure 6 to Figure 9. PSQM+ 0.4 0.72 0.78 DT-SQE 0.71 0.76 There are a few things to be noted here. First, the instrumental methods make almost no difference between the two cell re- placement methods: silence or repetition of last cell content (see g30_atm and g31_atm). But the subjective impression is that the repetition method gives better results than replacement method is useless, but that the transformation of PSQM values by silence. We do not know whether our data allow strong con- into MOS needs to be redesigned in the case of ATM cell loss. clusions to be drawn about the quality of cell replacement MOS estimated from PSQM+ is not quite linearly correlated methods. with subjective MOS either, but can be used as an approxima- tion. As was to be expected, dropping long packets leads to higher quality reductions than dropping single cells. PSQM+ and DT- For the right classification of the results Figure 10 shows the SQE properly register this fact, but PSQM has difficulties. auditory MOS values versus estimated MOS’ values gained by the instrumental approaches for all 53 conditions in the test. For The transformation of PSQM values to MOS values given in [4] this reason the instrumental speech quality values are mapped is useless. As can be seen from Figure 10, the estimated MOS into the MOS scale using a monotonous range of an optimal values range only from 3.95 to 4.05 for all files. MOS value 4 quadratic function. The five ATM conditions are drawn as corresponds to ‘good’ quality. This does not mean the PSQM black filled squares.
PSQM 0.0 PSQM 0.2 1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0 12345 12345 MOS MOS cell replacement, silence, g30_atm cell replacement, silence, g30_atm cell replacement, repetition, g31_atm cell replacement, repetition, g31_atm packet replacement, 3 cells, g32_atm packet replacement, 3 cells, g32_atm packet replacement, 6 cells, g33_atm packet replacement, 6 cells, g33_atm packet replacement, 6 cells, g34_atm packet replacement, 6 cells, g34_atm
Figure 6 PSQM 0.0 versus subjective MOS. 32-second files Figure 7 PSQM 0.2 versus subjective MOS. 32-second files
Telektronikk 3/4.1998 91 PSQM + 0.0 DT-SQE-MOS
5 5
4 4
3 3
2 2
1 1
0 0 12345 12345 MOS MOS cell replacement, silence, g30_atm cell replacement, silence, g30_atm cell replacement, repetition, g31_atm cell replacement, repetition, g31_atm packet replacement, 3 cells, g32_atm packet replacement, 3 cells, g32_atm packet replacement, 6 cells, g33_atm packet replacement, 6 cells, g33_atm packet replacement, 6 cells, g34_atm packet replacement, 6 cells, g34_atm
Figure 8 PSQM+0.0 versus subjective MOS. 32-second files Figure 9 DT-SQE versus subjective MOS. 32-second files
Instrumental MOS' Instrumental OMOS 5 5 DT_SQE PSQM (0.2)
4 4
3 ISDN - ISDN 3 ISDN - ISDN POTS - POTS POTS - POTS ISDN - POTS ISDN - POTS 2 GSM 2 GSM ATM ATM Simulation Simulation MNRU MNRU 1 1 1234 51234 5 Auditory MOS Auditory MOS Figure 10 Results gained by DT-SQE and PSQM(0.2) versus subjective MOS
4.2 Comparison of subjective MOS with cell loss vided by packets sent in active speech intervals, with an adjust- and packet loss rates ment for packets on the border of speech intervals). The defini- tions of active speech intervals and of silent intervals are taken Does the cell loss rate have any bearing on the subjective from PSQM, see [11]. The exact computation of sample loss impression of quality reduction, or does packet loss have more rate is described in [5]. to say? In Figure 11 the sample loss rate in active speech inter- vals is plotted against the subjective MOS. This corresponds to In Figure 12 the cell loss rate in active speech intervals is the packet loss rate (packets lost in active speech intervals di- plotted against subjective MOS. The cell loss rate refers only
92 Telektronikk 3/4.1998 MOS MOS 5 5
4 4
3 3
2 2
1 1
0 0 0 0.05 0.1 0.15 0.2 0 0.01 0.02 0.03 0.04 Packet loss rate Cell loss rate cell replacement, silence, g30_atm cell replacement, silence, g30_atm cell replacement, repetition, g31_atm cell replacement, repetition, g31_atm packet replacement, 3 cells, g32_atm packet replacement, 3 cells, g32_atm packet replacement, 6 cells, g33_atm packet replacement, 6 cells, g33_atm packet replacement, 6 cells, g34_atm packet replacement, 6 cells, g34_atm
Figure 11 Subjective MOS versus packet loss rate in active speech Figure 12 Subjective MOS versus cell loss rate in active speech
to cells lost on the ATM level (i.e. the cell loss trace), not those Conclusions on the instrumental methods lost inside packets on the ATM adaptation layer level. When comparing the different instrumental measures with the subjective MOS, PSQM+ and DT-SQE show a better correla- As was to be expected, the cell loss rate (inside active speech) tion than PSQM. Although these methods were not designed to has almost nothing to say on the speech quality. One has to take evaluate quality impairments caused by packet losses, PSQM+ into account how the protocols above the ATM layer pack cells and DT-SQE perform quite well for these conditions. into packets, and what happens when cells get lost. The instrumental methods possibly need to be improved to esti- One may also conclude that if the cell loss rate is maximally mate differences in speech quality, when different cell replace- 0.025, the quality of cell-impaired speech is ‘fair’, depending ment techniques are used. Subjective listening experience shows on the cell replacement method. that replacing lost cells by the content of the last received cell gives better quality than replacing lost cells by silence samples 5 Conclusions (the difference in quality is as large as from ‘fair’ to ‘good’), whereas all instrumental methods judge the quality of the two Conclusions for voice over ATM replacement methods to be roughly equal. Judging from eight speech samples heard by 24 listeners, the Limitations to this study quality of speech impaired by cell loss (not packet loss) is still fair, when the cell loss rate is about 0.025. Replacing lost cells This work has a few shortcomings, which could be the topic of by the contents of the last received cell gives a somewhat better further work. quality than replacing them by silence. • Cell delay and delay variation was not a part of the study. The sample loss rate (computed from the number of samples in • Only impairments to a single 64 kbit/s channel were investi- a lost packet) has more to say on speech quality than the cell gated. Encoding to lower bitrates, or multiplexing of several loss rate, which ignores packet sizes. channels into one ATM cell stream were not part of the study. • Reconstruction of lost packets or cells may have a beneficial The longer the packet size, the more serious the effect of cell effect on the voice quality. loss in the underlying ATM layer. This may as well hold true for IP-over-ATM connections. • Newer ATM switches give priority to CBR traffic such as traffic from voice connections. Thus it is expected that cell loss will be a very rare event in a well-dimensioned net. How- ever, when voice traffic creates a substantial part of the load, cell loss will play a role.
Telektronikk 3/4.1998 93 Acknowledgements 7 Helvik, B E, Stol, N. ATM traffic experiments : a laboratory study of service interaction, loss fairness and loss charac- We express our thanks to all members of EURESCOM project teristic. Trondheim, SINTEF-DELAB, 1995. (Report P603, especially to Bjarne E. Helvik for helpful discussions, and STF40 A95024.) to Michel Emons for providing us with PSQM evaluations of impaired speech files. Moreover, we thank Egil Aarstad for 8 ITU-T. B-ISDN ATM adaptation layer specification : type 1 configuring the ATM equipment. AAL. ITU-T Recommendation I.363.1, 1996.
9 ITU-T. Methods for subjective determination of trans- 6 References mission quality. ITU-T Recommendation P.800, 1996. 1 Alcatel 8643 ATGA User Manual. Zürich, Alcatel STR AG. 10 ITU-T. Modulated noise reference unit (MNRU). ITU-T Recommendation P.810, 1996. 2 Berger, J. Ein Ansatz zur instrumentellen Sprachqualitäts- abschätzung im Festnetz der Deutschen Telekom. (An 11 ITU-T Recommendation P.861. Objective quality measure- approach to instrumental speech quality estimation in the ment of telephone-band (300–3400 Hz) speech codecs, Deutsche Telekom’s fixed network). Workshop on quality 1996. assessment in speech, audio and image communication, ITG, EURASIP, Darmstadt, March 1996. (In German.) 12 Meky, M M, Saadawi, T N. Degradation effect of cell loss on speech quality over ATM networks. In: Mason, L and 3 EURESCOM. Quality of Service : Measurement Method Casaca, A (eds.). Broadband Communications. London, Selection. EURESCOM P603, Deliverable 1, QoS Chapman Hall, 1996, 259–270. overview, services and measurement methods : selected ser- vices and methods, 1997. 13 Myrstad, T. Test-svitsj for bredbånds ISDN (Test switch for broadband ISDN). Telektronikk, 88, (3), 63–65, 1992. (In 4 EURESCOM. Measurement method. EURESCOM P603, Norwegian.) Deliverable 2, 1997. 14 Wright, D J. Voice over ATM : An evaluation of imple- 5 EURESCOM. ATM Measurement Report. EURESCOM mentation alternatives. IEEE Communications Magazine, P603, Deliverable 2, Annex 1, 1997. 34, (5), 72–80, 1996. 6 Helvik, B E. Synthetic load generation for ATM traffic 15 Zwicker, E, Fastl, H. Psychoacoustics, facts and models. measurements. Telektronikk, 91, (2/3), 174–194, 1995 Berlin, Heidelberg, Springer-Verlag, 1990.
Abbreviations
AAL ATM Adaptation Layer ATGA ATM Traffic Generator Analyzer, Alcatel 8643 ATM Asynchronous Transfer Mode Jens Berger is Research Scientist in Quality and Acceptability of Tele-Services, a branch of Deutsche CBR Constant Bit Rate Telekom Berkom GmbH in Berlin. He started off in the area of acoustics and audio sound systems but is DT-SQE Deutsche Telekom – Speech Quality Estimation currently involved in national and international pro- HQTV High Quality Television jects in speech quality evaluation for telecommunica- tions. His Ph.D. thesis from 1998 focuses on instru- HSD High Speed Data mental (‘objective’) approaches for speech quality estimation. MOS Mean Opinion Score e-mail: [email protected] PSQM Perceptual Speech Quality Measure STG Synthetic Traffic Generator
Mechthild Stoer (35) is Research Scientist at Telenor Research and Development. She holds a Ph.D. degree from the University of Augsburg 1991 in the area of combinatorial optimization. Her interests are in network design and network quality issues. e-mail: [email protected]
94 Telektronikk 3/4.1998 Traffic from Mobility in Mobile Broadband Systems
FERNANDO J. VELEZ AND LUIS M. CORREIA
Models allowing the study of the influence of coverage At first glance, one could consider that the new calls traffic lin- distance and velocity on the supported traffic and on the ear density should be proportional to (1/R); however, this is new calls traffic linear density are examined, and results only valid for the static scenario, where there are no handovers. are obtained for typical scenarios in a Mobile Broadband In scenarios with mobility, the number of calls generated by System (MBS) with a linear coverage geometry. For systems handover increases as L decreases and the velocity increases, ξ without guard channels for handover, for a fixed bounding implying that n does not increase linearly with (1/R). For a ξ value for the blocking probability, the new calls traffic lin- given supported traffic, the increasing behaviour of n, asso- ear density was analyzed, increasing with the decrease of the ciated with the decrease of the coverage distance, corresponds maximum coverage distance, R, being upper limited by a to an increase of the cross-over rate η (number of handovers value which depends on the characteristics of the mobility per unit length) [4]. These facts originate higher probabilities scenario. However, call-dropping probability requirements for handover failure Phf and call dropping probabilities Pd with also need to be fulfilled, leading to a new calls traffic density the decrease of R. that only increases with the decrease of R down to an opti- mum value of R, and being lower for scenarios with higher The simple situation of homogeneous traffic (constant value of mobility. These optimum values of R are higher for scenar- new calls traffic in the whole service area) and linear coverage ios with higher and higher mobility, leading to limitations in geometry (where mobiles handover between the first and the system capacity, mainly for high mobility scenarios. In last cells, a typical geometry for roundabouts [5]) will be con- order to resolve these limitations, the use of guard channels sidered here as a first step to a more complicated (and closer to for handover is studied, particularly for high mobility reality) analysis. The objective of the design is to obtain values scenarios. For these scenarios one concludes that there is a for R that verify the requirements of system quality. These degradation in system capacity because, for the typical requirements consist of values lower than 1 – 2 % for the block- coverage distances foreseen for MBS, the new calls traffic ing probability Pb [6] and lower than 0.1 – 0.5 % for the call linear density is one order of magnitude below the values dropping probability [7]. obtained for the pedestrian scenario (where it is approxi- mately 15 Erlang/km), decreasing from 2.47 Erlang/km, in When no guard channels for handover are used the blocking the urban scenario, down to 0.84 Erlang/km, in the highway and the handover failure probabilities are equal, which strongly scenario, when two guard channels are used. limits system capacity for high handover failure probabilities, i.e. low coverage distances. If guard channels for handover are used, equations for blocking and handover failure probabilities 1 Introduction will be decoupled, and trade-offs can be used to improve system performance in the case of small cells. Mobile Broadband Systems will allow extending high data rates provided by the fixed broadband-ISDN to the cellular communi- Although MBS will be a multi-service system, providing multi- cations market, supporting high speed communications in high media mobile communications handling both bursty and con- mobility outdoor scenarios [1], leading to the use of millimetre stant bit-rate traffic, in this work only a single service will be wavebands. For such bands, shadowing from buildings is im- taken into account in order to simplify this first analysis of the portant, the propagation being mainly in line-of-sight. As a con- problem. A service with mean duration of 3 min is considered sequence, for urban scenarios the system will be based on a as an example. Anyway, this analysis will be very useful in the microcellular structure with cells confined to streets, having computation of multi-service aggregated traffic, because the dimensions in the order of a few hundred metres [2]; also for first step of this more general problem [8] consists of comput- main road scenarios, the use of microcells is foreseen. ing, for each type of traffic, the state probability as if it was the only kind of traffic in the system. The high mobility associated with the future Mobile Broadband System (MBS) yields a teletraffic analysis where both the new In Section 2, the influence of traffic from mobility on the micro- calls and the handover traffic should be considered simultane- cellular coverage distance is examined. Time parameters are ously. For systems where guard channels for handover are used, described and the handover probability is introduced. The char- this analysis is made assuming that handover traffic is Poisson acteristics of some main mobility scenarios are introduced and distributed [3], and that there is independence among the num- formulas for the cross-over rate are then obtained, from which ber of calls being served at each cell [4]. values for the average cross-over velocity are calculated. In sys- tems without guard channels for handover, for given values of One of the goals of system planning is the maximization of the the number of channels in each cell and of P , one can calculate new calls traffic in terms of the cells dimension, i.e. one is inter- b the corresponding supported traffic. Equations for the new calls ested in obtaining the cell coverage range that leads to a maxi- traffic and the traffic coming from handover are also obtained. mum new calls traffic density supported by the system. For the case of linear geometry, where mobiles travel randomly through cells with maximum coverage length R, total length L =2R, and base stations located in the centre, located end-to-end (Figure ξ 1), a new calls traffic per unit length n (or new calls traffic lin- ear density) is considered. To have an insight into the trade-offs 2R involved in the optimization procedure for MBS, the behaviour BS BS ξ I of n in terms of the maximum coverage distance of a cell R, needs to be studied for typical values of R, and for various mobility scenarios: static, pedestrian, urban, main roads and highways. Figure 1 Linear coverage geometry
Telektronikk 3/4.1998 95 Models for the computation of new calls traffic linear density Usually the service rate is assumed to be known for the service are presented, its dependence on R is highlighted and the result- or application under analysis, and the cross-over rate can be cal- ing limitations in system capacity imposed because of the high culated taking into account the distribution for velocities [3] mobility of terminals are discussed. 1 η = Vmax 2R · f(v)dv (2.3) In Section 3, the use of guard channels for handover is proposed 0 v as a prospective solution to overtake the limitations discussed in where v is the velocity and f is the velocity probability density Section 3 for the new calls traffic linear density. First, the function (note that in a linear geometry the total length of the assumptions made in the traffic analysis are described. Finally, cells is 2R). the trade-offs involved in the design are described and results are obtained for the supported traffic, the new calls traffic linear For a properly designed system, the new calls traffic density density and coverage distances that maximize it. increases as the coverage distance decreases, owing to the in- crease of the handover rate (mean number of handovers per call Finally, in Section 4, some conclusions are drawn on the in- when the probability of the handover failure is negligible) [4]; fluence of mobility on the new calls traffic linear density and this also causes the increase of handover failure and call dropp- its consequences on the optimization of MBS capacity. ing probabilities. The desired maximization of the new calls traffic linear density obeys to requirements of system quality, which consist of values lower than 1 – 2 % [6] for the blocking 2 Influence on the Optimum Micro- probability and lower than 0.1 – 0.5% for the call dropping cellular Coverage Distance probability [7]. An improvement in system performance can be achieved if guard channels are used for handover, but different 2.1 Traffic Requirements solutions are obtained depending on mobility scenarios and on the number of guard channels for handover, g: a total number of channels m = g + c is considered, where c is the number of In a linear coverage geometry, cells are placed end-to-end and channels to support both new and handover calls (Figure 3). mobiles can handover from a cell only to one of the two ad- jacent ones, Figure 2; a call comprises successive sessions τ , 1 The call dropping probability P is given by [4] τ , τ , ... in cells traversed by a mobile terminal, and its duration d 2 3 – • τ follows an exponential distribution whose mean is τ = 1 / µ i i µ τ Pd = PhPhf Ph (1 − Phf ) (2.4) [4], where is the service rate. The channel occupancy time c is the time spent by a user in communication prior to handover i=0 where i denotes the order of the handover and P is the hand- (or subsequent to handover) or call completion, which can also hf over failure probability. For small values of P , it can be app- be modelled by an exponential distribution with reasonable hf roximated by accuracy [9].
τ The cell dwell time h is the residing time of a mobile within a cell. Further assuming that the dwell time is exponentially dis- τ– η tributed with mean h = 1 / , then the channel occupancy time τ τ τ is c = min{ , h}, i.e. it is either the time spent in a cell before crossing the cell boundary if the call continues, or the time until the channel is relinquished [4]. As the minimum of two exponentially distributed random variables is also exponentially µ µ η Call distributed with parameter c = + , the mean channel occu- pancy time is given by completion µ 1 1 Blocked τ¯c = = (2.1) calls Pb µc µ + η
the probability of handover being given by New c λ η τ¯c calls i Ph =Prob{τ>τh} = = (2.2) µ + η τ¯h m
Handover λ g calls h Handover
Cell i Cell j Cell k failure Phf
τ Handover 2 τ τ η 2 3 calls τ τ 1 1 Figure 3 Model for the traffic in the case where the use of Figure 2 Dwell time and channel occupancy time guard channels for handover is considered (extracted from [3])
96 Telektronikk 3/4.1998 f(ν) η P = · P = ζ · P (2.5) d µ hf hf ∆ 2/(Vmax-Vmin) where ζ is the handover rate.
If guard channels for handover were not used, Phf would be equal to the blocking probability Pb [10], which imposes a strong limitation because Pb would be as low as Pd determines. V V V The use of guard channels for handover allows overtaking this min min max ν[ms-1] limitation because Pb and Phf will be decoupled [10]. In this case, depending on the coverage distance, the design is made by Figure 4 Velocity probability density function considering the traffic supported by m channels, from which g are guard channels [4].
The parameters involved in the design depend on the call gener- ation rate λ, the number of channels at each cell m, besides v Defining the average cross-over velocity η* as and µ. The simple situation of homogeneous traffic (constant value of new calls traffic in the whole service area) and linear η* = η ⋅ (2R) (2.9) coverage geometry (where mobiles handover between the first and the last cells, typical for circular geometry [5]) will be con- (η normalized to the cell length 2R) one obtains, for the scenar- sidered here as a first step to a more complicated (and closer to ios from Table 1, the values of Table 2. reality) analysis. The interest in defining this parameter is that it enables us to 2.2 Mobility Scenarios make explicit the dependence on R of some parameters to be defined later. The scenarios examined in the analysis are presented in Table 1, where a triangular distribution, with average Vav = (Vmax + Vmin) / 2 2.3 New Calls Traffic Linear Density without ∆ and deviation = (Vmax – Vmin) / 2, is considered for the veloc- Guard Channels for Handover ity [3] (Figure 4). For given values of m and Pb and a configuration which does The probability density function is given by not use guard channels for the handover calls, one can calculate ρ λ µ the corresponding traffic m = / by using the well known ρ ⎧ Erlang-B model [13]. The new calls traffic, n, and the traffic 1 ρ ⎨ ∆2 · [v − (Vav − ∆)] ,Vav − ∆ ≤ v ≤ Vav coming from handover, h, can then be obtained as [3] 1 f(v)= − 2 · [v − (Vav +∆)],Vav ≤ v ≤ Vav +∆ (2.6) µ 2µR ⎩ ∆ ρ = · ρ = · ρ 0, otherwise n η + µ m η ∗ +2µR m (2.10) which leads, when V , ∆ > 0, to the following cross-over rate η η∗ min ρ = · ρ = · ρ h η + µ m η ∗ +2µR m (2.11) 2R V +∆ η = (V +∆)· ln av 2 av ∆ Vav where the dependence on the cell length has been made explicit by introducing the cross-over average velocity. Note that V −1 ρ ρ ρ av m = n + h. −(Vav − ∆) · ln Vav − ∆ (2.7) Dividing (2.10) by (2R) one obtains a new calls traffic linear ∆ and when Vmin = 0 ( = Vav), to the limit density V 1 η = av · . (2.8) 2 · ln(2) (2R)
Table 1 Scenarios of mobility characteristics Table 2 Average cross-over velocity ⋅ –1 ∆ ⋅ –1 Scenario Vav[m s ] [m s ] η ⋅ -1 Static 0 0 Scenario *[m s ] Pedestrian 1 1 Pedestrian 0.72 Urban 10 10 Urban 7.21 Main roads 15 15 Main roads 10.82 Highways 22.5 12.5 Highways 21.21
Telektronikk 3/4.1998 97 y 2
With mobility lyzed. Given constant values of the blocking probability, one 1.6 obtains the graphs from Figure 5 for both situations (with and Without mobility without mobility).
ξ ρ µ η 1.2 Then, one could conclude that n is upper limited by ( m / *) [/m], which decreases with velocity. However, one should note that for a given R, x will be lower for higher velocities, and so y 0.8 will be larger, which partly compensates for the lower values of ρ µ η ( m / *) in equation (2.12).
0.4 However, call dropping probability restrictions also need to be fulfilled. Because Pb = Phf, the blocking probability should be computed according to (2.5) [3] and, because of that, the block- 0 012345678 ing probability will be a linear function of R with slope µ η x 2 (Pd)max / *. Figure 5 Normalized new calls linear traffic density y as So, the traffic supported by m channels will depend on R. For a function of x the considered scenarios an example is given in Figure 6 where: µ -1 m = 11, Pd = 0.5 %, = 1/3 min , and the values for the ∆ average velocity Vav and the velocity deviation are the ones presented in Table 1.
ρn µ ξn = = · ρm (2.12) As can be seen, the supported traffic will not be constant, rather 2R (η ∗ +2µR) having a decreasing behaviour with the decrease of R. In this ξ which can be normalized as follows case, the dependence of n on R will be as presented in Figure ξ 1 1 6b) for the three scenarios with higher mobility, i.e. for each y = n = = ρ · µ/η∗ 1+(µ(2R)/η∗) 1+x (2.13) scenario, there is an optimum value for the coverage distance m ξ which maximizes n. These maxima correspond to lower values where x = µ(2R) / η*. of R for the scenarios with lower mobility, being lower for the scenarios with higher mobility. For coverage distances lower ξ For the static scenario the new calls traffic linear density is than this optimum value, n decreases with the decrease of R. ξ ρ η n = m / (2R) because v = 0 and * = 0. Consequently, in a This is due to call dropping probability restrictions, which con- non-static scenario, if the contribution of mobility was not tradicts the behaviour expected with the simple analysis without considered, one would obtain y = 1/x. considering it. It is also worth noting that the new calls traffic linear density is much more sensitive to the mobility scenario From the operator’s point of view the objective is to maximize than the supported traffic because the former, besides the de- ξ ρ n, thus the dependence of this parameter on R should be ana- pendence on m, also depends on Ph.
ρ a)ξ [Erlang/km] b) m [Erlang] n 8 2.5 urban 7 pedestrian 2.1 main roads urban highways 6 main roads 1.7 highways 5 1.3 4 0.9 3
2 0.5 0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0 R [km] R [km] Figure 6 Traffic for m = 11 channels and the design made according to the call dropping probability restrictions a) Traffic supported, b) New calls traffic linear density
98 Telektronikk 3/4.1998 Pd 0.6 3 Traffic from Mobility with Guard Channels for Handover 0.5 pedestrian urban 3.1 Blocking and Handover Failure Probabilities 0.4 main roads highways Considering the use of guard channels for handovers when no 0.3 queueing of new or handover calls is performed, the blocking and handover failure probabilities are given by [3]. This is 0.2 assuming that the handover traffic can be approximated by a Poisson process [8] and that the new calls traffic is also Poisson distributed, which is valid for a number of users in a cell much 0.1 larger than the supported traffic. 0 0 0.2 0.4 0.6 0.8 1.0 + k (ρ + ρ )c c g ρh R [km] P = n h k=c k! b −1 ( + )k + k (3.1) c ρn ρh +(ρ + ρ )c c g ρh Figure 7 Call-dropping probability for g = 0 k=0 k! n h k=c k!
ρg (ρ + ρ )c h n h (c+g)! Phf = (3.2) c−1 ( + )k c+g ρk ρn ρh +(ρ + ρ )c h k=0 k! n h k=c k!
ρ ρ ρ where n is the new calls traffic and h is the handover traffic. While Pb(R) is almost constant with R (the right part of the ρ ρ ρ ρ The traffic supported by m channels is then m,g = n + h. curves in Figure 8 (example for g = 1)), Phf(R) increases with R (since it was obtained according to (2.5) and η depends on R, as The new calls traffic and the traffic coming from handover can in (2.8) and (2.9)) (the left part of the curves). Thus, a break- ρ ρ then be easily obtained as in Section 3.3, replacing m by m,g point exists at the intersection of both curves, where ρ ρ [3]. The new calls traffic linear density is then given by Phf(R) = Pb(R), and for values of R lower than this breakpoint ρ 1 µ the curves have then an appreciable slope. One can observe that ξ = n = · ρ n 2R 1+2µR/η∗ η∗ m,g (3.3) the supported traffic decreases as the velocity of the associated scenario increases, mainly in the zone of the curves limited by handover failure. It can also be seen that the breakpoint occurs ρ µ ξ µ η For a fixed m,g and a given , n is upper bounded by / *, for increasing values of R for faster and faster mobiles. decreasing with the increase of the cross-over velocity, and hav- ρ ing a variation with R of the type 1 / (1 + x), where x is directly As could be expected, when g increases, Pb(R) decreases. ρ ρ proportional to R and inversely proportional to the cross-over However, as m,g(R) is the minimum between Pb(R) and ρ average velocity. Phf(R), this decrease is only effective for the part of the curves ρ ρ where Pb(R) is lower than Phf(R). The challenge in the design for low values of R is finding values for g that, for a given m, 3.2 Supported Traffic ρ ρ both maximize Phf and keep it lower than Pb, mainly for the scenarios with high mobility (urban, main roads and highways). The examples given here were obtained for the conditions men- ρ tioned before. For g = 0, using the supported traffic m,g that verifies Pb = 2 %, which does not depend on R, one obtains val- 3.3 New Calls Traffic Linear Density ues for the new calls traffic linear density that increase with the decrease of the coverage distance. However, the corresponding For given values of m and g, we can then obtain the curves for ξ call-dropping probability constraints associated with (2.5) and the new calls traffic linear density n(R) according to (3.3). Fig- with Pb being equal to Phf (Figure 7) are only fulfilled in the ures 9 and 10 show these curves for g = 1, 2 (the latter only for ρ pedestrian scenario and only for R > 300 m. A way to resolve the higher mobility scenarios). For the pedestrian case, as m,g this limitation without drastically decreasing the new calls is almost constant for all the range of R, we basically observe traffic linear density, is the use of guard channels for handover. that it follows the behaviour of the ratio at the right member ξ from (2.3). For the three scenarios with higher mobility, n(R) From an operator’s point of view, in order to achieve MBS pro- presents maxima, depending on the velocity and on g. These visional coverage distances, approximately in the range maxima occur for distances lower than the breakpoints, for ξ 100 – 350 m [2], one intends to increase n while R decreases. design purposes corresponding to optimum values of R, Ropt In order to obtain results for the supported traffic the procedure (Table 3), and agreeing with the provisional values for MBS was the following: taking Pd = 0.5 %, (2.5) was used to get a and also with the need to use larger cell lengths for high mobil- value for Phf. With this Phf value and with Pb = 2 %, (3.1) and ity scenarios owing to the cost associated with signalling [12]. ρ (3.2) were solved separately for the supported traffic m,g (using One can also see that the breakpoints occur for lower coverage ρ ρ (2.10) and (2.11)), and the respective values, Pb and Phf, were distances as g increases. obtained. In order to cope with both probability requirements, the minimum of these two must be taken.
Telektronikk 3/4.1998 99 ρ [Erlang] ξ m,g n [Erlang/km] 5.5 25
pedestrian 5.0 20 urban main roads 4.5 15 highways 4.0 pedestrian 10 3.5 urban main roads 5 3.0 highways
2.5 0 0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0 R [km] R [km] Figure 8 Traffic supported by m = 11 channels with g = 1 Figure 9 New calls traffic linear density, for m = 11 and g = 1
ξ [Erlang/km] ξ n n [Erlang/km] 2.5 0.85
urban 0.8 2.1 main roads highways 0.75 1.7 0.7
1.3 0.65 g = 0 g = 1 0.6 0.9 g = 2 0.55 g = 3 0.5 0.5 0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0 R [km] R [km] Fig 11 Figure 10 New calls traffic linear density, Figure 11 New calls traffic linear density for m = 11 and g = 2 for m = 11 and g = 0, 1, 2 and 3, in the highway scenario
ξ Figure 11 presents n(R) for the highway scenario for several 4 Conclusions values of g. One observes an improved new calls traffic linear density for g = 2 for 160 < R < 610 m, the maximum A microcellular communications system without guard channels ξ n = 0.84 Erlang/km being obtained for Ropt = 475 m. An for handover, with a linear coverage geometry, was first ana- improvement of the new calls traffic linear density when g = 3 lyzed. Models to compute the supported traffic and the new exists only for 130 < R < 160 m, the maximum being obtained ξ for Ropt = 130 m, n = 0.81 Erlang/km.
It is noticeable that the use of guard channels makes a differ- Table 3 Approximate values for Ropt and maximum values for ξ ence in system performance, especially for high speed scenar- n with m = 11 ios, where it allows us to overcome the problems associated ξ with handover failure constraints. For these scenarios and for Ropt [m] n [/km] the typical coverage distances in MBS, the new calls traffic lin- ear density is one order of magnitude below the values obtained Scenarios g = 1 g = 2 g = 1 g = 2 for the pedestrian scenario (where it is approximately 10 – 15 Urban 125 150 2.40 2.47 Erlang/km), decreasing from 2.47 Erlang/km, in the urban scenario, down to 0.84 Erlang/km, in the highway scenario. Main roads 175 250 1.60 1.65 Highways 375 475 0.82 0.84
100 Telektronikk 3/4.1998 calls traffic linear density as a function of velocity and cell 5 Sidi, M, Starobinski, D. New call blocking versus handoff length were examined. For a fixed blocking probability, one blocking in cellular networks. Wireless Networks, 3, (I), verifies that the new calls traffic linear density, which was used 1997, 15–27. as a measure to system capacity, is upper limited by the average cross-over velocity of the associated scenario, having lower val- 6 Correia, L M et al. Report on design rules for cell layout. ues for scenarios with high mobility, and varying differently RACE Deliverable R2067/IST/2.2.3/DS/P/044.b1, RACE depending on the average cross-over velocity. However, for Central Office, Brussels, Belgium, 1994. actual scenarios the call-dropping probability constraints are not fulfilled in such conditions. Thus, in order to simultaneously 7 ITU-T. Networks grade of service parameters and target verify the constraints for the blocking and the handover failure values for circuit-switched public land mobile services. probabilities, it is necessary to consider lower blocking proba- ITU-T Recommendation E.771, Geneva, 1996. bilities which leads to a decrease in the new calls traffic linear density with the decrease of R for such configurations. Feasible 8 Ross, K, Tsang, D. Teletraffic engineering for product-form values for the new calls traffic linear density were obtained, circuit-switched networks. Adv. Applied Probability, 22, where there are optimum values for the coverage distance that 1990, 657–675. maximize it. These maxima correspond to lower values of the coverage distance for scenarios with lower mobility, being 9 Guérin, R. Channel occupancy time distribution in a cellular lower for scenarios with higher mobility. radio system. IEEE Transactions on Vehicular Technology, 36, (3), 1987, 89–99. In order to overtake the limitation associated with the case where no guard channels are used, configurations that use guard 10 Hong, D, Rappaport, S. Traffic model and performance channels for handover were analyzed. Results were obtained for analysis for cellular mobile radio telephone systems with different values of the number of guard channels g. One con- prioritized and non-prioritized hand-off procedures. IEEE cludes that, for the coverage distances foreseen for MBS, higher Transactions on Vehicular Technology, 35, (3), 1986, optimum values for the new calls traffic linear density are ob- 77–92. tained for g = 2. However, for the highways scenario the corre- sponding coverage distances are larger than the foreseen cell 11 Yacoub, M D. Foundations of mobile radio engineering. lengths. It has also been verified that there is a degradation in CRC Press, Boca Raton, Florida, USA, 1993. system capacity, measured in Erlang/km, for higher and higher mobility scenarios. 12 Silva, N, Brázio, J. Reduction in location management sig- nalling by the use of subpopulation-tailored location area sizes. In: Proc. of ICUPC’ 96 – International Conference on Acknowledgements Universal Personal Communications, Cambridge, Mas- sachusetts, USA, 1996, 602–606. Fruitful conversations are acknowledged with Prof. João Luis Sobrinho of IST and with Prof. Vladimir Gligic from University of Beira Interior. The authors would like to express their sincere thanks to the Department of Electromechanical Engineering of the University of Beira Interior, for providing the computational resources that enabled part of the work of Fernando Velez to be successfully accomplished. Fernando J. Velez (28) is Lecturer at the Dept. of Electromechanical Engineering at Univ. of Beira Inte- References rior, Covilhã, Portugal, and PhD student in Electrical and Computer Engineering (ECE) at IST, Technical 1 Fernandes, L. Developing a system concept and technolo- University of Lisbon. He holds a degree in ECE from gies for mobile broadband communications. IEEE Personal IST and received his MSc in ECE in 1996. His re- Communications Magazine, 2, (1), 1995, 54–59. search interests include cellular planning, cost/rev- enue performance of mobile communication systems and traffic/mobility aspects. He is currently working 2 Velez, F, Correia, L M. Optimization criteria for cellular on multi-service traffic in mobile broadband systems. planning of mobile broadband systems in linear and urban e-mail: [email protected] coverages. In: Proc. ACTS Mobile Communications Summit, Aalborg, Denmark, Oct. 1997, 199–205.
3 Chlebus, E, Ludwin, W. Is handoff traffic really Pois- Proc. of ICUPC’ 95 – IEEE International Con- Luis M. Correia received his PhD in 1991, and is cur- sonean? In: rently a Professor at IST in the area of Telecommuni- ference on Universal Personal Communications, Tokyo, cations. His research work is focused on wave prop- Japan, Nov. 1995, 348–353. agation, antenna design, traffic modelling and cellu- lar planning in mobile communications. He has been 4 Jabbari, B. Teletraffic aspects of evolving and next-genera- involved in European research, currently in ACTS, as tion wireless communication networks. IEEE Personal well as Chairman in COST 259. He has also served Communications Magazine, 3, (6), 1996, 4–9. as Auditor/Evaluator in the ACTS and ESPRIT pro- grammes. e-mail: [email protected]
Telektronikk 3/4.1998 101 The haves and have-nots – The distribution of ICTs in Norwegian households
RICH LING
• The interactive communication technologies (ICT) haves – • The greatest attitudinal strife between the haves and the defined here as those who consume far more than the have-nots is found among mature adults aged 35 – 55. The average number of electronic devices – are dominated by haves and the have-nots in this age group disagree about those in the younger age groups while the have-nots are the role of ICT in daily life, the ability to integrate it into largely over 55. Analysis shows that the group of elderly the home and in their general perceptions of systems such have-nots is becoming smaller and smaller as the younger as the Internet. groups carry their consumption practices with them as they age. There seems to be a shift in the consumption Introduction of technology between those born before and after the Second World War. A great deal of interest has been directed towards the question • Income is perhaps the single most important factor of the penetration of ICTs into the private sphere. There is a determining one’s status as a have or a have-not. concern that we are developing a dichotomous society of those who have and those who do not have these devices. The afflu- • Attitudinal factors and one’s ‘ideological’ orientation ence of Norway has led portions of the population to become towards ICT are important in determining one’s con- quite advanced consumers of electronic devices while others lag sumption of these devices. behind in this activity. In this paper we focus on two social groups with quite opposite life experiences, namely the ICT haves and have-nots in Norway. As one will see in this paper Number of respondents differences in one’s consumption of ICTs reveal significant dif- ferences in age, education, income, the use of media along with 140 differences in attitude on a range of issues.
120 In Figure 1 we have displayed the distribution of devices from 1 100 one of the two databases used to analyse this phenomenon . It is the cases at the high and the low end of this distribution that are Have nots 2 80 defined as the haves and the have-nots . 60 Age makes a big difference 40 Haves One of the first things that strikes the reader is variation in con- 20 sumption of electronic devices by age. In Figure 2 we see that the mean number of devices in the home generally declines as 0 the age of the interviewee increases. There are, however, 012141620246810 18 nuances in this picture. One can see that interviewees between Number of devices 18 and 35 reported fewer devices than the youngest respon- Fig.1 dents. This represents the fact that they had recently moved Figure 1 Distribution of electronic devices among the Norwegian population out of the parents’ home, had a less stable economy and a more nomadic existence. Respondents between 35 and 55 years of age had a relatively large number of devices while, as we will see below, those over 55 reported the ownership of fewer and fewer devices. Mean number of devices 14
12 1 The ‘Telenor R&D database’ analysis included the following devices in the inventory of ICTs: PC, printer, modem, CD- 10 ROM, more than one TV, cable, satellite dish, video, CD player, game console, telefax, copy machine, mobile tele- 8 phone, more than one telephone handset, more than one tele- phone line, telephone answering machine, video camera. The Statistics Norway analysis included the following inventory of 6 items: PC, CD-ROM, modem, Internet, CD player, one or more TVs, video player, cassette player, record player. 4 Finally, the analysis of generational effects is based on the following inventory of items: one or more TVs, video player, 2 game console, PC. 2 The distribution has the general form of a normal curve. The 0 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 mean is slightly higher than 7 items and the first standard Age deviation is 3.7 units above and below the mean. The groups Fig.2 are roughly the same size but since the definition units are discrete devices it was impossible to establish two equally Figure 2 Mean number of devices sized groups.
102 Telektronikk 3/4.1998 Figure 3 shows the distribution of the haves, have-nots and the Percent of age group remaining population by age3. The distribution is divided into 100 five age-based groups. These include youth (younger than 18 Have nots 90 years of age), young adults (18 – 35), established adults (35 – Normal 55), ‘pre-retirement’ adults (55 – 65), and retirees (older than 80 Haves 65). The youth and the retirees are groups defined by their rela- tion to age-bounded social institutions, namely the school sys- 70 tem and work life. The figure shows a rapid drop in the percent 60 of haves as this group migrates into the young adult phase. 50
The remaining three groups are defined empirically. The young 40 adults are in the process of establishing families and the pursuit of careers. The relatively small number of haves in this group 30 points to a certain economic pressure here. The percent of haves 20 was higher for respondents between 35 and 55. This is a conse- quence of the fact that the established adults have generally 10 overcome the first hurdles of establishing themselves and enjoy 0 a greater stability than the previous group. The established 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 adults have the income with which to purchase the various Young Mature Pre-retirement ICTs and they are also often pressured to purchase by teenage Youth Retired children living at home. adults adults adults
As the respondents moved into their mid fifties, i.e. the ‘pre-re- Figure 3 Percent of age group tirement adults’ group and the retirees, we see that the percent of have-nots rises rapidly. This trend continues through the remainder of the distribution. These two groups seem to have a different relationship to ICT than those who are younger. Sev- eral elements may conspire to form this situation. Many work and home related ICTs have arrived on the scene only after this Percent group was well established in their careers. Accommodating them in their everyday routines may have been more difficult 90 1994 than for those who were introduced to the technologies at a 80 younger age. These two groups do not have the same domestic 1995 pressure to participate in the ICT revolution as teenage children 70 1996 have generally moved out of the home. Finally, the youth, 1997 young adults and the established adults were born after the Sec- 60 ond World War. The ‘pre-retirement adults’ group and the 50 retirees were born before this dramatic historical event. While the former grew up during an extended period of prosperity and 40 belief in technology, the latter grew up during the end of the depression and the hardships of the war. This early life ex- 30 perience may have had an effect on their willingness to use 20 money on relatively expensive devices. 10 This argumentation suggests that as the established adults migrate into the older age groups they will carry with them their 0 taste for ICTs. To examine this we have compared the percent 10 15 2025 30 35 40 45 50 55 60 65 70 75 Age of have-nots using a standard list of ICT holdings between 1994 Source: Statistics Norway and 1997. The data shown in Figure 4 indicates that the high percentage of have-nots associated with the elderly is being Figure 4 Aging of the have nots pushed into older and older age groups. The data shows, for example, that while in 1994 the point at which 20 % of the elderly were have-nots was at age 50, this had risen to 56 by 1997. duction of ever new technologies such as Digital Video Display This data indicates that one can expect the elderly have-nots to (DVD) and Web TV will moderate the reduction of have-nots become a smaller and smaller portion of society as those with among the elderly. ICT experience move into this age group. However, as we will see below, limited income in this group along with the intro- Income is a key factor
3 The data in Figures 3 and 4 is represented by a seven-year Changing gears now, we will examine the five groups described moving average in order to flatten out the more extreme above for differences in income, education and media use. The swings in the data. data dictates, however, that in the following analysis we must
Telektronikk 3/4.1998 103 define the haves and the have-nots age-group by age-group as Income per year (NKr x1000) opposed to the a single definition across all groups as has been 4 600 used in the analysis up to now . Have nots 500 Haves The analysis indicates that a major factor determining one’s status as a have or a have-not is the household’s income. Figure 400 5 shows that there were significant income differences within all five age groups. Even though the age differences were 300 slightly more pronounced in the lower age groups, they were quite considerable in every case. It is often the case that gender, 200 educational/career choice and income are related factors as women generally have lower incomes. In our analysis, however, 100 we examine household income that tends to disguise this differ- ence. 0 Youth Young Mature Pre-retirement Retired**5 Another variable that showed widespread differences between adults** adults** adults** the haves and the have-nots was education. In Figure 6 we see the percent of respondents who had completed a college educa- Figure 5 Income per year5 tion in each of the age groups6. The data shows that there were strongly significant differences between the haves and the have- nots in the last three groups. Our data, in fact, showed that it was difficult to find have-nots with a college education.
Another aspect of the differences between the haves and the Percent have-nots is the use of other media. The data here shows that 50 the haves among the retirees are oriented towards reading while Have nots the haves among the younger age groups are higher than aver- Haves 40 age telephone users. In Figure 7 we see the mean number of minutes per day re- 30 ported for reading newspapers. The interesting point here is the significant difference between the haves and the have-nots 20 among the retirees. While all other age groups were quite simi- lar on this point, there was a considerable difference in this age 10 group. This difference is likely a consequence of the educational differences pointed out above.
0 At the other end of the age spectrum we find a difference in the Young Mature Pre-retirement Retired number of telephone calls made per day by the haves and the adults adults** adults have-nots. While there were differences across all age groups Figure 6 Percent of respondents with higher education here, it was among the youth and the young adults that we found the significant differences between groups.
Attitudinal differences: Established adults show the greatest differences Minutes per day In the data we also gathered information on the respondents’ 120 attitudes towards ICT. Our analysis shows that these were as Have nots strong an indicator of one’s have/have-not status as the more 100 Haves traditional background variables examined above. Thus, we 80
60 4 In practical terms, the use of a single definition across all 40 ages would mean that there are too few haves among the retirees just as there are too few have-nots among the youth. 20 Thus, haves and have-nots were defined as those above and below the respective standard deviation for each age group. 0 5 The following symbols are used in the remaining figures: Young Mature Pre-retirement + sig. = 0.062, * sig. < 0.05, ** sig < 0.001. Youth adults adults adults Retired* 6 The youngest age group was excluded from this analysis as Figure 7 Time per day spent on reading newspapers they were still in the educational system.
104 Telektronikk 3/4.1998 were able to not only point out demographic differences be- Calls per day tween the groups, but also examine some of the attitudinal 4 differences between the groups. Specifically we included a Have nots series of 11 questions in the questionnaire examining their atti- Haves tudes toward the Internet and towards PCs at home. The ques- 3 tions covered various aspects of ICT such as the perceived importance of the Internet in future work situations, the degree to which the PC belongs in the home and general glosses of the 2 Internet.
The first point to be made in this analysis is that with the ex- 1 ception of the established adults there is little disagreement be- tween the haves and the have-nots in regards to the role of the Internet. By contrast, the data shows a complex of differences 0 between the haves and the have-nots who are over 35 years of Young* Mature Pre-retirement Youth* Retired age with regard to the PC. adults adults adults Figure 8 Use of telephones Somewhat tellingly, have nots among the retirees felt that the PC was too complex for use in the home and that it represented a disturbing element in the home. In Figures 9 and 10 one can see the relative weight that each of the age groups placed on these items on a five-point scale. Figure 9 in particular shows a dramatic increase in the degree to which the have-nots among the retirees felt that the PC was too complex for use in the Agree 5 home. Figure 10 points to the degree that the respondents felt Have nots that the PC was a disturbing element in the home. In this case Haves there were significant differences between the haves and the 4 have-nots for both the retirees and the established adults.
The attitudinal data indicates that the ideological battleground 3 over ICT in the home is among the established adults. It is in this age group that one finds the most extensive differences be- tween the haves and the have-nots. There was disagreement as 2 to the extent of ICT’s impact in society. This can be seen in the disagreement over the statements “Eventually everybody will buy a PC”, “Knowledge of how to use a PC will be important in Disagree 1 school and work”, “The Internet has importance for working Young Mature Pre-retirement life” and “The Internet is a fad”. In addition, there was disagree- Youth adults adults adults* Retired** ment as to the ability to integrate PCs into the home “The PC can disturb domestic life” as well as the general understanding Figure 9 The PC is too complex for use in the home of the Internet “The Internet is only porn and violence”. Figure 11 shows the relative disagreement between the haves and the have-nots on these items for the established adults.
Conclusion
In conclusion, our analysis has shown that the haves are domi- Agree 5 nated by those in their teens and those in their 30s, 40s and early 50s. The have-nots are largely over 55. Analysis shows that the Have nots Haves elderly have-nots are being replaced by younger cohorts who 4 are bringing their consumption patterns with them into the ‘pre- retirement adults’ and retirement phases of life. The analysis also shows that the orientation of those born before and after the 3 Second World War seems to be dramatically different in regard to the consumption of ICT. While the former are comfortable with the consumption of a variety of items the latter are more 2 grudging in their use of these devices.
Income and education are important clarifying factors when Disagree 1 understanding one’s interest in the consumption of ICT. The Young Mature Pre-retirement higher the income and the higher the educational level the more Youth Retired** likely one is to be among the haves. Educational level is a par- adults adults* adults ticularly important factor for the older respondents. Figure 10 The PC is a disturbing element in the home
Telektronikk 3/4.1998 105 Eventually everybody will buy a PC* Finally we find that one’s ‘ideological’ orientation towards ICT is important in determining one’s consumption of these devices. With the exception of the established adults aged 35 – 55, the Knowledge of how to use a PC ideological orientation towards the PC is not problematic. Either will be important in school and work* the age group is uniformly comfortable or uniformly uncomfort- The Internet has importance able with this technology. By contrast, the age groups over 35 for the working life* are far more divided in their attitude towards the Internet.
The Internet is fad* When looking at the specific age groups, the data shows an ideological divide between the haves and the have-nots among The PC can disturb domestic life** the established adults aged 35 – 55. Here one sees disagreement HaveHave nots nots as to the role of ICT in daily life, its integration into the home HavesHaves and in the general perception of ICT services. The Internet is only porn and violence*
1 2345 Method Disagree Agree The data for this paper comes from two main sources. The main Figure 11 Disagreement among established adults analysis focuses on Statistisk sentralbyrå’s (Statistics Norway) Mediebruksundersøkelse from 1997 and from data gathered by Telenor R&D in 1997 on PC ownership. The former source included 2000 interviews carried out at four points spread throughout the year. The population of this survey was a repre- sentative sample of persons aged 9 to 79. The latter includes 1000 interviews carried out in the spring of 1997 wherein the respondents were a representative sample of Norwegians rang- ing from 13 to 91 years old. Both questionnaires gathered infor- mation on the consumption of domestic electronic devices. The R&D survey had a somewhat more extensive list of devices than did the Statistics Norway survey. The R&D survey in- cluded items such as TV, radio, video machine, PC, CD player, fax machine, copier, telephone answering machine and the like.
Rich Ling (44) is a sociologist working at Telenor R&D, Kjeller. He received his Ph.D. in sociology from the University of Colorado, Boulder. Since coming to Norway he has worked at the Resource Study Group founded by Jergan Randers and has been a partner in the consulting firm Ressurskon- sult. For the past five years he has worked with Telenor and has been active in researching issues associated with new information technology and society. e-mail: [email protected]
106 Telektronikk 3/4.1998
Kaleidoscope
Telektronikk 3/4.1998 107 108 Telektronikk 3/4.1998 NORDTEL – 80 years of Nordic co-operation in the telecommunications area EINAR UTVIK AND VILLY TOFT
1917 – 1997 in chronological perspective systematic and long-sighted perspective, i.a. by establishing permanent groups/committees to work more continuously in the The telecommunications administrations of Sweden, Denmark periods between the telecommunications conferences. To study and Norway started co-operation on an occasional basis as well this proposal more in detail, the meeting established an expert as bilateral discussions already in 1858, but the more permanent group under the chairmanship of Torsten Larsson, Sweden. The Scandinavian, and some years later, the genuinely Nordic task of this group was to work out a detailed proposal for the telecommunications co-operation, started in 1917. During organisation and form of the further co-operation. The final World War I the governments of the three countries had started proposal of this expert group was presented to the thirty-fifth a co-operation on foreign policy based on their common inter- meeting of NT, in Reykjavik in 1971. ests as neutral states. The telecommunications administrations were also involved in this co-operation, with a particular view The debate on this proposal at the conference resulted in the to making the censorship of the telegraph and telephone traffic signing of the ‘General agreement on Nordic co-operation in the more effective. Such censorship was considered necessary in telecommunications area’ in Reykjavik on 27 August 1971. As order to prevent the transmission of information that could provided in this agreement, the Nordic Telecommunications endanger the security of Scandinavian merchant vessels, and Conferences (NT) should continue to be held every second year it was also desirable to keep the propaganda of the belligerent alternating between the Nordic countries and pursuant to the powers within reasonable limits. same rules as those previously applied.
During such a government conference in Copenhagen in To supervise the NT activities between the conferences, a steer- December 1917, the directors general of the three telecommuni- ing committee, NST, was established, consisting of the five cations administrations took the opportunity to conduct internal directors general and under the chairmanship of the director discussions on matters purely related to telecommunications, who was to host the next NT conference. Three co-ordination and this meeting has later been considered as the first Scandi- committees were also created to assist the NST in the manage- navian telecommunications conference. Such discussions were ment of the different working groups: pursued on a regular basis, and from the eighth meeting in • NTD – for operational and tariff questions Stockholm in October/November 1924, there was also participa- tion by the Finnish and Icelandic telecommunications admini- • NTR – for technical questions related to radio communica- strations. Consequently, from this conference onwards, the tions co-operation can be considered as being genuinely Nordic. The • NTT – for technical questions related to telecommunications reason why Finland and Iceland were invited to this particular (excluding radio communications). conference was that the agenda included discussion of proposals that were to be presented to the International Telegraph Confer- At the thirty-sixth NT conference in Rold Storkro, Denmark, in ence in Paris in 1925. June 1973, three additional committees were established: The Nordic telecommunications conferences (NT) were held with irregular intervals until the twenty-first meeting in Helsinki in December 1947. Subsequently, the conferences took place regularly every second year, with the exception of the extra- ordinary conferences in 1954 and 1964.
The conference venues alternated between the different coun- tries in such a way that the meetings were held in Iceland in the years x1, in Denmark in the years x3, in Sweden in the years x5, FOTO: BESSING FOTO in Finland in the years x7 and in Norway in the years x9. The duration of the conferences was normally 4 – 5 working days.
To a certain extent, visits to industrial undertakings or sight- seeing tours were organized in between the regular conference sessions. In addition to trade-related and cultural inspiration, these supplementary activities offered possibilities for more informal, collegiate discussions between the participants.
During the telecommunications conferences, the questions on the agenda normally covered points handled by the three stand- ing committees: the Traffic and Telecommunications Com- mittee, the Technical Committee and the Radiocommunications Committee.
At the thirty-fourth meeting of NT in Kabelvåg, Norway, in June 1969, the Swedish delegation proposed an extension of the Nordic telecommunications co-operation. The main reason for this proposal was the substantial increase in the number of items on the agenda of the conferences, and it was considered neces- sary that the activities of the conferences were treated in a more The cable ship “Peter Faber” with submarine cable plough
Telektronikk 3/4.1998 109 • NTSK – Nordic Telecommunications Satellite Committee NT decided to create a working group to handle this matter and to present a proposal for a new agreement at the next NT con- • NKTK – Nordic Cable TV Committee ference. • TS – Tanum Board (for the operation of the common satellite earth station in Tanum, Sweden). The new agreement, ‘Agreement and guidelines for Nordic co- operation in the telecommunications area’, was signed at the Seventeen working groups presented their reports to this con- thirty-ninth NT conference in Stavanger, Norway, on 31 May ference, but already four years later, in 1977, the number of 1979. working groups had increased to thirty. In this agreement it was stated that “the permanently organised At that time – during the thirty-eighth NT conference in Savon- Nordic co-operation takes place in Nordic telecommunications linna, Finland – the Danish delegation proposed that the rules conferences (NORDTEL), in committees and working groups”. for the future Nordic telecommunications co-operation should be revised. The principal reason for this was that even if im- The following committees were created to supervise the work- portant results had been achieved, it was a fact that the co-oper- ing groups within their respective areas of competence: ation required great resources, and the main idea behind the pro- • NM – the committee for marketing, network planning and tar- posal was to rationalise the co-operation activities through a iff questions partial delegation of the management responsibility of NST to the co-ordination committees, i.a. by giving these the compe- • NT – the committee for technical questions related to tence of establishing and disbanding working groups. In this telecommunications (excluding radio communications) way, NST would have more time to treat fundamental and • NR – the committee for technical questions related to radio principal, future-oriented questions and problems. communications. This committee was also the supervisory
FOTO: NN
NORDTEL meeting, Faroe Islands, June 1993. Walking on the beach after inauguration of the CANTAT-3 terminal station at Tjörnavik
110 Telektronikk 3/4.1998 FOTO: ULF OWENEDE body of NTSK (Nordic telecommunications satellite com- mittee) • ND – the committee for data communication • TS – the Tanum Board.
This agreement provided a well suited framework for the acti- vities during the 10-year period from 1979 to 1989.
At the end of this decade, however, it became clear that the gradual liberalisation and growing competition in the telecom- munications area in Europe made it increasingly difficult for the working groups to collaborate closely on commercial and also on technical questions. It was therefore considered necessary to revise the current agreement, and after long and thorough prepa- rations in a working group, a new agreement was signed at the NORDTEL meeting in Tórshavn on 8 June 1993. NORDTEL meeting, Gothenburg, August 1995 This new agreement implied the disbanding of the former NM, NT and NR committees (ND had already been disbanded some time earlier).
Two new groups were established under NORDTEL: NS, the committee for network strategy, and KM, the contact group for market development. There was also agreement on maintaining NP, the committee for network planning, which had been split off from NM in 1989 as a separate committee.
About four months earlier, on 17 February 1993 in Copenhagen, the five managing directors of the NORDTEL member organi- sations had also established a new trade association for telecom- contract with the association. At the same time the board de- munications operators in the Nordic countries, NTOB, ‘Nordic cided to consolidate the association’s secretarial activities in Telecommunications Operators’ Trade Association’. The asso- Copenhagen and to appoint Mr Villy Toft, Tele Danmark, as ciation was open for membership by all telecommunications director of NTOB for a period of nine months. network operators in the Nordic market and provided a separate legal identity for the trade co-operation (association registered During these nine months a revaluation of NTOB’s tasks and in Copenhagen, subject to relevant Danish legislation). organisation was to take place. In this revaluation, the stated wish of NORDTEL to consider the possibility of merging the According to its statutes, the association should also have a sep- two organisations and their activities should be taken into arate office in Brussels, Belgium. This office was established in account. August 1993, at the same time as the first director of NTOB, Mr Svein Tenningås, Televerket, Norway, took up his duties which As a result of this revaluation, the NORDTEL members (the mainly consisted of discharging NTOB’s functions in Brussels. managing directors of Telia, Telenor, Tele Danmark, Telecom Finland and P&T Iceland) agreed at their meeting in Porvoo, Within NORDTEL, it proved increasingly difficult to continue Finland, on 13 May 1997, to merge NORDTEL with the four the activities within KM, the contact group for market develop- year old trade association, NTOB, with a clear recommendation ment, because of the steadily growing competition. This called of using the name NORDTEL, and not NTOB, for the merged for another revision of the NORDTEL agreement, and the re- organisation. vised agreement was signed at the NORDTEL meeting in Gothenburg, Sweden, on 8 August 1995. The new agreement In conformity with this decision, the general assembly of NTOB implied the disbanding of KM while the two other committees agreed to the proposed merger at its meeting in Copenhagen on • NS – the committee for network strategy, and 28 May 1997 and amended the statutes to provide the formal basis for the new, merged organisation. The subsequent meeting • NP – the committee for network planning of the NTOB board, which was held the same day, decided to continued their activities. accept Tele Danmark’s offer to make Villy Toft available for the position as NTOB director for another year. After the establishment of the trade association, NTOB, the majority of the members set up their own, individual office in As a final consequence of this reorganisation of the Nordic Brussels. Taking into account the fact that in addition to Den- telecommunications co-operation, the members of the former mark, Sweden and Finland had become members of the Euro- NORDTEL organisation, at their meeting in Oslo on 4 Decem- pean Union, the NTOB board decided in July 1996 to close ber 1997, decided to terminate the agreement signed in Gothen- down the association’s office in Brussels. The office was closed burg in August 1995 and to abolish the old organisation. on 1 September 1996, the date of termination of the director’s
Telektronikk 3/4.1998 111 Overall survey stations required so huge investments that it would have been very difficult for the individual Nordic telecom operator Initially, the Nordic collaboration was characterised by ques- (administration) solely to meet such expenses. By taking this tions relating to tariff matters. The main purpose of these efforts up as a joint activity all the Nordic telecom operators (admin- was to keep tariffs for inter-Nordic telecommunication traffic as istrations) became involved in the satellite communications at low as possible without disregarding requirements for a satis- an early stage. factory economical result of the telecommunication activities. • Collaboration on joint representation in the satellite organisa- tion INTELSAT. Each of the individual Nordic telecom After World War II the collaboration was extended to include operators (administrations) was too small to obtain a seat on joint economical activities as well. A number of successful and INTELSAT’s Board of Governors, but by forming a group, a positive achievements emanating from this collaboration could seat in the Board of Governors and in the sub-committees was be mentioned, in particular from the 1970s and 1980s when the ensured. co-operative activities were at their peak: • Co-operation within European and international organisations • Co-operation on joint planning, extension and exploitation of such as ITU, CEPT, ETSI, EURESCOM, ETNO etc. This co- inter-Nordic traffic connections and connections from the operation was established very early (ref. joint preparations Nordic countries to other countries, often organised as a pool for the International Telegraph Conference in Paris, 1925) co-operation. This co-operation resulted in significant eco- and has since characterised and strengthened the participation nomical savings for all the Nordic telecom operators. of the Nordic telecom operators (administrations) in the inter- • Joint development and establishment of NMT – the Nordic national telecommunications organisations. Today this co- Mobile Telephone system – which turned out to be a great operation constitutes an important part of the activities in the success and which laid the foundation of the outstanding new Nordic trade association. position of the Nordic countries within mobile telecommuni- cations. Summing up, the tradition of a constructive co-operation among the major incumbent Nordic telecommunication operators has • Joint development and establishment of the public Nordic lasted for nearly a century. data network (DATEX). • Joint establishment and operation of the satellite earth stations Despite great differences between the Nordic countries in topo- in Tanum, Eik and Ågesta for the INTELSAT, INMARSAT graphy, population density and differences in the national orga- and EUTELSAT systems respectively. The first satellite earth nisation of the telecom enterprises, common cultural, political
FOTO: BILDCENTER - FARSTA
NORDTEL (NTOB) Board meeting, Stockholm, October 1997
112 Telektronikk 3/4.1998 FOTO: FRIDOLFS BILDCENTER and demographic characteristics have formed the basis of signif- icant mutual interest also within the telecommunication area.
A number of the achievements have been quite tangible (ref. the examples mentioned above), whereas others have had their impact through strategic political perspectives and may there- fore have been less visible to those parties not directly involved in the questions concerned.
From an overall point of view it is fair to state that this long- standing collaboration has contributed to place the Nordic countries among the most innovative, dynamic and highly de- veloped telecommunication markets of the world.
Status and perspectives for the future
In 1998 the overall political and market situation of the telecom- munication business is characterised by the following: • A number of new services • Increased liberalisation Using the mobile phone in Nordic surroundings • Political measures aiming at promoting competition (inten- tion to have more players in the market) • Changed regulatory schemes with the declared aim to have real competition as quick as possible, in some cases by intro- ducing discriminating conditions for the incumbent (national) dominant operators • Internationalisation – for business enterprises (our major cus- tomers) as well as for telecom operators • Expanding market for traffic handling and for service pro- vision • Convergence between the IT market, the entertainment/cable TV industry and the conventional telecommunications market • Technological convergence between the IT, telecommunica- tions and the entertainment industry’s technologies (‘ICT technology’)
• International alliances. Einar Utvik (58) is deputy director of Regulatory and International Affairs in Telenor AS. He is responsible Taking this environment into account the Nordic telecom co- for general and policy matters related to international operation will therefore focus more on overall interests common organisations. to all professional players and less on matters related to opera- e-mail: [email protected] tional questions which are of significance to the commercial and competitive situation of the individual enterprises.
Examples of such common areas of interests could be: • Increased market volumes • Harmonised business conditions (regulation) • That customers do not perceive the range of products offered as non-coherent and unnecessarily complicated to use Villy Toft (64) is deputy director of Tele Danmark’s department for International Relations and Stan- • Synergy oriented co-operation in parallel with open mutual dardisation. Since 1992 his main area has been co- competition (‘coopetition’). ordination of Tele Danmarks activities within ITU, ETSI, TRAC and other international consortia and The main challenge for the new NORDTEL organisation in the fora. Villy Toft has since September 1996 been years to come will be to further develop the activities within director of the Nordic Telecom Operators Business these areas of common interest and to increase the membership Association, NORDTEL (former “NORTEB”). in order to create a solid basis for continued Nordic co-opera- e-mail: [email protected] tion in the telecommunications field.
Telektronikk 3/4.1998 113